1 Introduction

English words such as lab, exam, or gator are the product of a truncation process that is called clipping (Plag 2003: 116). In the literature on linguistic morphology, clipping is commonly seen as a minor word formation process that is relatively unpredictable (Durkin 2009: 116) or even outside the confines of morphology (Haspelmath & Sims 2010: 40). These positions are grounded in the observation that clipping is highly variable. Bauer (1994: 40) states that “there is no way to predict how much of a word will be clipped off in clipping, nor even which end of the word will be clipped off”. This view is shared by Don (2014: 27), who argues that “clippings can result from deletion of just any part of the word, and there does not seem to be a clear pattern”. The following examples illustrate that variability.

Clipping may involve the truncation of the end of a word (sis), the beginning (gator), or both end and beginning (fridge). The phonological material that is clipped off and deleted may be part of a syllable (shroom), an entire syllable (sis), or several syllables (gator, fridge). The clipped form may consist of one syllable (sis) or more syllables (gator). It may end in a consonant (sis) or in a vowel (bro). The same word may be clipped in different ways (bro, bruv). Clippings mostly instantiate nouns (sis), but also other word classes, including adjectives (legit). There is by now a growing literature on clipping that tries to account for this variability (Lappe 2007; Jamet 2009; Berg 2011; Alber & Arndt-Lappe 2012; Arndt-Lappe 2018) and that challenges the view of clipping as an unsystematic, unpredictable phenomenon. In this paper, we extend that work on the basis of new data and the use of multivariate statistics, with the goal of constructing a probabilistic, multifactorial model of the clipping process. That model should generate predictions about the form of a clipped word on the basis of the non-truncated source word, and it should allow us to explain why speakers produce that specific form. As has recently been pointed out by Arndt-Lappe (2018: 146), this aim has up to this point not been pursued with enough effort: “A general problem in assessing the predictability of output structures of truncatory processes is that, in much of the pertinent literature, the degree of predictability is not in the focus of investigation.”

Previous research into the clipping process (Tournier 1985; Antoine 2000; López Rúa 2002; Jamet 2009; Balnat 2012, amongst others) has produced sizable collections of clipped forms, as well as several empirical generalizations. The position that clipping is largely unsystematic (Bauer 1994; Durkin 2009) is still common in current textbooks on word formation (Don 2014), but it has not gone unchallenged. In particular, studies that approach the phenomenon on the basis of quantitative evidence report probabilistic tendencies that suggest that clipping is at least to some extent predictable. In an early contribution, Tournier (1985: 298) notes the predominance of end-clipped forms in English words that are listed in the Concise Oxford Dictionary. Corroborating this finding, Jamet (2009: 17) analyzes a dataset of 290 clipped forms taken from Wikipedia, finding that more than 90% of them are nominal and that 75% involve end-clipping, rather than front-clipping or clipping of both the beginning and the end of the source word. Tournier (1985: 303) also finds that a majority of English clipped forms end in a consonant. Tournier (1985: 305) further argues that competing functional pressures give rise to the shape of clipped forms. While it would be economical for speakers to truncate as much material as possible from a source word, clipped forms need to retain enough substance to be recognizable. Put simply, clippings need to achieve a balance between the principle of least effort (Zipf 1949) and the principle of maximized recognizability.

Other early research has addressed the meaning of clipped forms. Two conflicting positions can be found. On one side, Adams (1973: 135) states that clipping is “the process by which a word of two or more syllables (usually a noun) is shortened without a change in the function taking place”. Contradicting this view, there are accounts that view clippings either as informal or as belonging to specialized discourse. The informal nature of many clippings, which is illustrated by cig from cigarette, has been observed already in Kreidler (1979). Empirically, Adams’ assessment is contradicted by clippings such as bro or merch, which have undergone lexicalization processes (Brinton & Traugott 2005; Hilpert 2019) that have rendered them less than fully synonymous with their respective source forms brother and merchandise. The clipped form bro serves as an intimate form of address, while the clipped form merch tends to refer to a very specific type of merchandise, namely clothing items that are purchased by the fans of pop artists and other celebrities. Plag (2003: 121) points out that clippings can differ from their non-truncated sources by conveying the speaker’s familiarity with the denotation of the clipped form. Many clipped forms refer to highly specific referents that are only known to a limited community. As such, clipped forms function sociolinguistically as a marker of in-group identity. Rock musicians know that the clipped form Strat refers to a specific guitar model (Stratocaster), which can be plugged into an amp (amplifier) that is connected to a cab (loudspeaker cabinet). However, the community specificity of clippings is not a general feature. Many clipped forms have gained wider currency, so that a mic can be identified as a microphone not only by musicians.

A proposal that is relevant for our account of English clippings has been proposed by López Rúa (2002), who addresses clippings and other shortening phenomena in English from the perspective of prototype theory and radial networks (Lakoff 1987). López Rúa argues that clippings, initialisms, abbreviations, and blends can be analyzed as related phenomena. Her account, which is based on a database that includes 195 clippings (López Rúa 2002: 34), identifies prototypical schemas for each individual process, including clipping. The idea that clipping can be understood in terms of several different schemata will be of central concern for the analysis in this paper.

As was mentioned above, Lappe (2007) offers a pioneering analysis of clippings and other shortening phenomena in English that provides strong evidence against the view of clipping as an unpredictable phenomenon. The analysis is based on 2259 truncated names and 1158 suffixed and unsuffixed clippings (2007: 59). It is shown that a number of phonological factors, including stress, syllable structure, and consonant clusters can be used to account for systematic correspondences between shortened forms and their source words (2007: 164). Couching her analysis in an optimality-theoretic framework, Lappe (2007: 168) views the clipping process as the product of two interacting types of constraints, namely faithfulness constraints, which bias speakers against outcomes that differ strongly from their sources, and phonological markedness constraints, which bias speakers against phonologically unusual outcomes. Faithfulness constraints (cf. McCarthy 2011: 13) include the tendency to preserve the stressed syllable of the source word in the outcome ('doctor > doc). Markedness constraints include the tendency for shortening to result in monosyllabic words, since these represent the unmarked case in English. Lappe (2007: 182–183) proposes different constraint rankings for monosyllabic and disyllabic clippings and shows that these rankings successfully address the variability that can be observed in the data. The present paper builds on these results, but it also differs from Lappe’s approach in several respects. First, the present analysis restricts itself to the analysis of clipped common words to which no further material has been added. This means that hypocoristic names such as Andy (< Andrew) or Betty (< Elizabeth), diminutive forms such as ciggy (< cigarette) or comfy (< comfortable), o-suffixed words such as pervo (< pervert), or blends such as brunch (< breakfast + lunch) are not taken into account. Second, the present analysis draws on a new large database of more than 2000 English clippings. Third, it differs in its methodological approach, specifically its use of Hierarchical Configural Frequency Analysis (HCFA).

Another study that is close in spirit to the present study is offered by Berg (2011), who examines a database of 935 English clippings in order to determine the factors that govern speakers’ choices between end-clipping and front-clipping. While end-clipping is generally the more frequent choice, Berg (2011: 9) uses logistic regression modeling to study the impact of different factors. The analysis shows several main effects and interaction effects. Source words with non-initial stress show a relatively greater likelihood of front-clipping (e.g. gator < alligator). Also, shorter words have relatively greater odds of front-clipping. First names are front-clipped more often than common nouns, especially when they have non-initial stress (e.g. Beth < Elizabeth, Tricia < Patricia). Berg (2011: 12) proposes a discourse-pragmatic explanation for the observed differences between first names and common nouns, suggesting that front-clipping poses a greater risk to the recoverability of a source word (cf. Nooteboom 1981), and that first names, due to their contextual grounding, are more easily recovered than common nouns. This explains speakers’ general preference for end-clipping and why first names deviate from this tendency more often than common nouns.

Alber & Arndt-Lappe (2012) propose a conceptual distinction between templatic truncation and subtractive truncation, which is relevant for the discussion in this paper. In their words, “[i]n templatic truncation, the size of the derived form is predictable; in subtractive truncation, predictability holds for the size of the truncated material. Prediction of form in both cases normally involves reference to higher-level phonological categories such as the syllable or metrical units such as the foot” (Alber & Arndt-Lappe 2012: 290). Templatic truncation thus allows the analyst to make generalizations about the shape of clipped forms. In phenomena that instantiate subtractive truncation, generalizations can be stated about the material that is taken away from a source word. As will be described in more detail in section 2, we have annotated a large database of attested English clippings for variables that capture characteristics of both the source words and the clipped forms, so that we are able to analyze different patterns of clipping in terms of either templatic or subtractive truncation.

In other work that assumes a cross-linguistic perspective on clipping, Alber (2010) and Arndt-Lappe (2018) study name truncation and clipping across languages such as Italian, German, and English in order to come to terms with the variability that shortening processes exhibit both within individual languages and across different languages. Instead of analyzing clipping as a unified phenomenon, different systematic truncation patterns are identified, whose formal characteristics are shown to correspond to different functional purposes. With regard to hypocoristics, strikingly systematic correspondences can be observed between non-truncated sources and their truncated targets (Arndt-Lappe 2018: 149). In Italian, truncated names tend to preserve either the first syllable (Albi < Alberto) or the main stress (Betta < Elisabetta). Truncated names with an i-suffix such as Albi overwhelmingly preserve the first syllable, while truncated names with a reduplicated consonant (Pippo < Filippo) show a strong preference for the preservation of the main stress. On the basis of these findings, Arndt-Lappe (2018: 145) argues, contra Haspelmath & Sims (2010), that it would be premature to exclude clipping from the study of grammatical morphology.

Our approach in this paper further aligns with Gries (2006), who focuses on English blends (brunch < breakfast, lunch) and complex clippings (sysadmin < system, administrator) with the aim of uncovering both the underlying systematic patterns that govern these processes and the cognitive factors that motivate them in the first place. Amongst other things, Gries (2006: 555) finds that the recognition points of source words play a more important role for blends than for complex clippings. Blends also tend to recruit source words that are relatively more similar to one another in terms of their edit distance. The quantitative evidence leads Gries to conclude that the two processes are much more systematic than has been previously acknowledged.

To sum up the preceding paragraphs, a substantial body of research weakens the traditional view of clipping as unsystematic and unpredictable. A number of regularities have been detected, and the literature has identified several cognitive, discourse-pragmatic, and phonological factors that are implicated in clipping. Existing work has also laid important groundwork by drawing conceptual distinctions such as the contrast between templatic and subtractive truncation. The present paper will go beyond the current state by drawing on a larger database, by applying a method that up to now has not been used for the analysis of clipping, and by adopting Construction Grammar as a theoretical framework that is particularly suited for the study of clipped forms.

The paper is structured as follows. Section 2 discusses our data and methods. With regard to the data, a database of more than 2000 English clippings has been compiled from various sources. Each entry has been annotated for a range of variables that capture different structural criteria of the clipping and its respective source word. We then discuss the basic principles of Hierarchical Configural Frequency Analysis (Krauth & Lienert 1973; Gries 2008), which we adopt as the analytical method in order to model the clipping process. Section 3 presents the results of our analysis, which identify fourteen structural patterns that occur more frequently than expected in the database. We further discuss another outcome of the analysis, namely two patterns that are significantly underrepresented. In section 4, we discuss our results in the light of earlier research on clipping, and specifically the role of functional pressures that shape the clipping process. Section 5 concludes the paper and outlines how we plan to follow up our results in future work.

3 Results

The calculations for the present analysis were performed with HCFA 3.2 (Gries 2004), a script written for the statistical software package R (R Development Core Team 2020).³ The analysis establishes that there are several significant interactions between the variables that were described in section 2.1.3. Section 3.1 discusses fourteen types that appear in the database more often than would be expected. Section 3.2 inverts the perspective and discusses two configurations that are conspicuously absent from the database. In the terminology of the HCFA, these configurations are called anti-types.

4 Discussion

The results that were obtained allow us to reaffirm the claim that the clipping process is not completely unpredictable (Lappe 2007; Jamet 2009; Berg 2011). To put this assessment into perspective, the 14 types that we presented in section 3.1 account for 770 of the 2272 clippings in the database (33.9%). There certainly is a lot of variability and even unpredictability, but a quantitative analysis of authentic data brings out several regular tendencies that reflect cognitive, discourse-pragmatic, and prosodic factors (cf. Arndt-Lappe 2018). The types that were identified by the HCFA are sensitive to segmental, suprasegmental and morphological structure, and they reflect patterns that balance recoverability and economy. This point has been articulated in several ways in the literature, and our results serve as another piece of empirical evidence that corroborates it. The following paragraphs outline three points that are raised by our results and that go beyond the issue of non-predictability.

First, our analysis suggests that it is useful to understand clipping in English as a process that involves several smaller-scale generalizations. Put simply, there are several ways to clip words in English. Each of these ways can be thought of as a schema that draws on the same general strategy of truncation, but does so in different ways. The types that we have discussed under labels such as merch, romcom, or hydro have the status of separate generalizations that speakers can use to generate new, similar clippings. This state of affairs can be usefully captured by adopting the perspective of Construction Grammar (Goldberg 2019), especially as it pertains to phenomena in morphology (Booij 2013; 2019). If clippings exhibit a large variety of forms, that is because there are several different constructional schemas that yield different outcomes, not because the clipping process as such would be unconstrained. In other words, clipping may appear to be unpredictable if it is viewed at the most general level, but at more specific levels, there is order. The importance of lower-level generalizations has been pointed out by Langacker (1999: 106), who argues that grammatical patterns that are specific have several processing advantages over patterns that are more abstract. The more specific a pattern is, the more features an individual instance will share with the schema that licenses it. This is especially important in the case of clippings, which incur a risk of miscommunication unless the clipping reliably allows the hearer to recover the source word. It is thus motivated that clippings should be organized around lower-level generalizations, which is a conclusion that resonates with Booij’s (2019: 393) assessment that “grammar and lexicon are not separated, and that the architecture of the grammar should be conceived as a multidimensional network of relations between syntactic and morphological constructions”.

Second, based on our results we argue that clipping patterns are usefully understood as generalizations that map a source word to a shortened structure. This point takes us back to Alber & Arndt-Lappe (2012) and their distinction between templatic and subtractive truncation. Templatic truncation is oriented towards the clipped form, specifically the prosodic form that is the output of the clipping process. Subtractive truncation, by contrast, focuses on the source word and specifies which parts are truncated in the clipping process. A clipping type such as the type romcom can be usefully described in terms of templatic truncation. Clippings of this type conform to a template that has two closed syllables and carries initial stress. The speaker’s knowledge of the template further specifies that the source word is a compound rather than a non-compound word, and that only initial material of the two component words is maintained. A parallel argument can be made for the clipping type that yields forms such as stats, nugs, and deets. The prosodic template of this clipping type is monosyllabic and it ends in a consonant cluster that contains the plural suffix -s. The source word needs to be a plural noun, which is affected by truncation in such a way that the stem is end-clipped and the plural suffix is retained. Crucially, analysis is not clipped to ans, and neither is progress clipped to progs. These potential clippings bear a superficial resemblance to the previously mentioned examples, but since they do not derive from plural forms, they do not arise. The relation between a clipping template and its source word can be seen as a paradigmatic link between linguistic forms that are mutually associated. Recent work that combines Construction Grammar with Relational Morphology has elaborated the concept of ‘sister links’ (Audring 2019) as an organizational principle of linguistic knowledge. Instead of thinking of clipping patterns as cognitive schemas that subsume aspects of the clipping template and the characteristics of the source word, an analysis in terms of sister links would rely solely on the connections between the two, without positing an overarching generalization. While it would be beyond the scope of this paper to work out such an account in detail, we do believe that this approach is highly promising for the analysis of clipping.

As a third point, we submit that the types and antitypes that were identified by the HCFA give rise to hypotheses that can be tested empirically. The issues that were raised in the preceding paragraphs allow us to formulate precise claims that are open to experimental investigation. Generally speaking, we hypothesize that in a forced-choice task in which one option corresponds to a type that has been identified by the HCFA, while the other option represents an anti-type, the type should be preferred. For example, if participants are presented with the word probation and the two hypothetical clippings pro and prob, we hypothesize to see a preference for the latter, which instantiates the type croc that was discussed above. Given the word renovation and the options reno and renov, we expect to see a preference for the former, which corresponds to the type hydro that the HCFA has identified. There are more intricate predictions that our results allow us to make. Recall that the romcom derives from a compound source word, while this is not the case for the type hydro. We therefore predict that given the source word high-potential, the choice between the two alternative clippings high-po and high-pot should be biased towards the latter. However, given the source word hypotension and the phonologically identical alternatives hypo and hypot, we expect a preference for the former. In other words, we hypothesize that the same outcome structures are judged differently depending on the purported source word of the clipping.

Summing up this section, we believe that the points we have raised illustrate that clipping is a far cry from the trivial phenomenon as which it is sometimes portrayed. Developing a theoretically satisfying account of the kind of variation that is exhibited by clippings is a veritable challenge for different linguistic frameworks.

5 Concluding remarks

This paper has presented a newly-compiled database of more than 2000 English clippings, and it has discussed the results of a multivariate analysis that has identified several patterns that reflect regularities in the way speakers of English create shortened words. We have made the case that English clipping is best understood as a set of lower-level generalizations, which allow us to better perceive the reliable tendencies in what at first glance appears to be a bewildering, unconstrained process. While this point has been argued in similar form by López Rúa (2002), we believe that our analysis adds new empirical support and greater depth of description. We have further argued that clipping patterns contain information that relates to both the input and the output of the clipping process, and we have pointed to Relational Morphology as a useful theoretical framework for further analysis. We sketched possibilities to use experimental techniques in order to engage with the predictions that can be derived from our empirical results.

To conclude this paper, we acknowledge that we have had to sidestep several issues with regard to clipping that would have merited closer attention. For one thing, our analysis has not been able to take into account the frequencies with which the clippings in our database appear in authentic language use. The clipping types that we have discussed have been identified on the basis of their type frequency alone. A more thorough investigation of the productivity of clipping patterns would have to draw on relevant corpus-based measures (Baayen 2009), which is something we hope to do in future work. It is also clear that analogy is a phenomenon that cannot be ignored in the context of clipping, especially with regard to highly frequent forms that are not part of a more general pattern, as for example the clipping bro (< brother), which, according to our results, actually instantiates a clipping anti-type. Monosyllabic clippings with final vowels are underrepresented in our database, so if speakers nonetheless create new clippings of this kind, this might receive a more compelling explanation in terms of analogy. Clippings further have fascinating semantic and socio-linguistic aspects, which we have not been able to work into the present analysis. Very often, the use of a clipping conveys the idea of shared common ground and a tight social bond between speaker and hearer. In future work, we further hope to contribute to the growing literature that investigates clipping from a cross-linguistic perspective (Nübling 2001; Alber and Arndt-Lappe 2012; Balnat 2012), specifically as regards a contrast between clippings in English and in French. There is thus no shortage of questions that can be asked about clippings and that deserve our close attention.

Supplementary Files

The data on which our analysis is based is available as a supplementary file.

Funding Information

The research reported on in this paper was funded by the Swiss National Science Foundation under grant 100015_188788 (“Clipping in a cross-linguistic perspective”, PI Martin Hilpert).

Competing Interests

The authors have no competing interests to declare.