At -ing and on -ing: Comparing two sentential complements of the verb work

Mark Kaunisto & Juhani Rudanko
University of Tampere 

Abstract

1. Introduction

One interesting characteristic of the use of the verb work is the variety of complement patterns that the verb takes. This paper examines two types of sentential complements of the verb, exemplified in sentences in (1a-b):

In both (1a) and (1b) the matrix verb work selects a preposition, at in (1a) and on in (1b), and an immediately following -ing clause as its complement, with the -ing clause being a gerund. In (1a) the complement clause is what is here termed an at -ing complement and in (1b) it is an on -ing complement. It is assumed here that the gerundial complements are sentential, with their own understood subjects. Such understood subjects make it possible to represent the argument structures of the lower predicates in (1a-b) in a straightforward fashion.

The syntactic structures of (1a) and (1b) are similar in another important way. In both types of sentence, the matrix verb work assigns a semantic role to its subject, and the prepositional -ing complements of work therefore involve control. More specifically, they involve subject control in both (1a) and (1b). Eschewing elaborate projections, of the type of AgrS and AgrO, the structures may be represented as in (2), with the symbol PRO (see Chomsky 1985: 119–131) representing the understood subject:

The representation in (2) makes use of the traditional notion of a nominal clause, which is simply a clause dominated by an NP node.

In addition to the two gerundial patterns, the verb work also selects other patterns in English. Among them are to infinitives, as in “Corporate management often works to keep the resolution off the annual meeting agenda” (COHA, 2000, MAG), and nonsentential patterns with the prepositions at and on, as in She worked at the tax office and They worked on the plan. Such additional patterns deserve attention, but for practical reasons the present investigation focuses on the two gerundial patterns. The two patterns are similar syntactically in that they consist of a preposition and a gerund, and they are also fairly similar in meaning (see Section 3). In view of the similarities of the two patterns, it is of interest to compare them in more detail with data from large electronic corpora. This article offers such a comparison. Section 2 of the paper first offers a general survey of the use of the two constructions by observing the numbers of tokens in two large electronic corpora. Section 3 provides a discussion on the semantic aspects of the constructions, also paying attention to the earlier work on the use of the patterns. Finally, Section 4 looks into other possibly distinguishing characteristics of the two patterns that can be investigated by quantitative methods.

2. Survey of data from British and American English

The Corpus of Historical American English, COHA, is here used as the source of data for investigating the history of the at -ing and on -ing patterns selected by the matrix verb work. The choice of this database was made because of the large size of this corpus and because of its balanced nature. To collect relevant tokens the following basic search strings were used: [work].[v*] at [v?g*] and [work].[v*] on [v?g*], which derived tokens of different inflected forms of the verb work followed by the prepositions at and on and a gerundial form of any verb. In order to ensure that relevant tokens could be retrieved which had intervening words between either the matrix verb and the preposition or between the preposition and the lower-level gerund, separate searches were conducted with wildcard asterisks between the elements. Eventually searches were made of tokens with up to six intervening words between the matrix verb and the preposition, and zero, one or two intervening words between the preposition and the gerund (relevant tokens were not found with more than one word between the preposition and the gerund). Irrelevant tokens observed by manual inspection were excluded from further analysis.

 Table 1 gives information about the frequencies of the two types of gerundial complements of work in each decade of the corpus. Information is also included on the size of the subcorpus for each decade, given in millions of words. Normalized frequencies per million words are given in parentheses.

This historical survey yields a number of noteworthy descriptive results. It shows that both patterns were very rare in the first few decades of COHA, and that it was the at ‑ing pattern that was the first to emerge in noticeable numbers. For its part, the on ‑ing pattern was conspicuous by its absence during the entire nineteenth century. In the early decades of the twentieth century the on ‑ing pattern continued to be very rare, and much less frequent than the at ‑ing construction. However, from the 1940s onwards it began to show an upward trend, with the trend being in keeping with the general spread of gerundial complements in fairly recent English (see e.g. Rohdenburg 2006, Vosberg 2009). Both patterns have become more frequent in the last few decades, but even more spectacularly, the on ‑ing pattern has surpassed the at ‑ing pattern in frequency by a fairly clear margin in the two most recent decades of the corpus. The changes over time are visualized in Figure 1 below, with the normalized frequencies of the tokens from periods of thirty years collapsed together. The figure illustrates the increase of both patterns, as well as recent surge in the number of tokens of the on ‑ing pattern.

In view of the rather dramatic rise of the on -ing pattern in very recent English, it is of interest here also to consult a second corpus, with data from recent American English. The 100-million-word Corpus of American Soap Operas was selected here, in order to consider data that may approximate current spoken American English, at least to some extent. Both variants are found in this corpus in plentiful numbers. Here is an illustration of each:

Overall, there are as many as 389 relevant tokens of work on ‑ing in the Soap Opera Corpus, which is 3.9 per million. On the other hand, the number of at ‑ing complements of work is 85, which is 0.8 per million. The difference in frequency between the two patterns is therefore even more lopsided in favor of the on ‑ing pattern in this corpus than in the most recent decades of COHA.

There is no corpus of British English currently available that would precisely match the last two decades of COHA or the Soap Opera corpus with respect to its structure or the dates of its contents. The British National Corpus (BNC) is probably the best one can do here, even admitting that a portion of the data in it comes from slightly earlier. Be that as it may, both variants are certainly found in the BNC. Here is an illustration of each:

Overall, there are 90 relevant tokens of the at -ing pattern and 119 tokens of the on -ing pattern in the BNC. Comparing these figures with the findings based on the last two decades of COHA and taking into account that a major segment of the BNC predates the 1990s, it appears that British English shows a distribution of the two types of complement that is fairly similar to the American English data. There does not appear to be a colonial lead in the frequency in the spread of the on -ing pattern with work.

3. Comparing the meanings of the two constructions

It was stated in the introduction that the meanings of the two gerundial patterns are fairly similar. This section offers a discussion of the issue. We might begin by considering these two pairs of recent examples:

Even when the investigator is armed with real data, it is not easy to discern a difference in meaning between the two patterns. The task is compounded when it is noticed that the two patterns can be found under the same sense in some major dictionaries. Some dictionary entries may be considered here. In the online edition of the Oxford English Dictionary (OED), the third major meaning category of the verb work is given as “To labor, toil,” and under this major sense there is sense 12, which is phrased as follows:

Among the illustrations of sense 12 there is one token of what is here called the at ‑ing complement:

There is no example of an on ‑ing complement given under the sense, but there is an example of an on NP complement of the verb, and, more importantly, the preposition on is listed alongside of at in the types of complements selected by the verb when it has sense 12.

As for the Collins COBUILD dictionary, the following subentry of work from the 1987 edition is helpful and may be quoted in full:

In the Collins COBUILD entry there is again an emphasis on at/on NP complements, instead of the sentential patterns, but one of the examples illustrates a gerundial pattern, and the sense definitions in Collins COBUILD, especially the first of them, are worth bearing in mind. Similar paraphrases for work at and work on can be found in other present-day English dictionaries, for example Macmillan English Dictionary (2002) and Longman Dictionary of Contemporary English (2003). In the Macmillan English Dictionary (2002), separate entries are provided for both work at and work on as phrasal verbs taking gerundial complements. Work at (s.v.) is paraphrased as ‘try hard to develop or improve something’, with the example If she works at improving her game, she could be a champion, and the definition for work on (s.v.) reads ‘spend time producing or improving something’, with the example He’ll have to work on getting fit before the game.

The failure of major dictionaries to make out a distinction between the two patterns in terms of meaning is a reflection of the degree to which the meanings are similar. Even so, it is appropriate to consider the task of teasing the two patterns apart, even on general theoretical grounds. Here we may recall Bolinger’s famous dictum: “a difference in syntactic form always spells a difference in meaning” (Bolinger 1968: 124). It is recalled that in light of the data from COHA the at ‑ing pattern is the older of the two, and it would be a strange situation if the on ‑ing pattern had arisen alongside of the at ‑ing pattern with precisely the same meaning. It is the purpose of this section to make a start on the task of comparing the two patterns, with an emphasis on a qualitative comparison, and on the task of differentiating their meanings in the system of English predicate complementation.

Any attempt to tease the meanings of the two patterns apart must be grounded in comparing the concrete difference between them, that is, in the fact that in one case the gerund is introduced by the preposition at and that in the other case it is introduced by the preposition on. It should of course be recognized that in both constructions the sentential constituent consisting of at or on and the following gerund is a complement of the verb work, and therefore it cannot be expected that the meaning of the preposition in the construction would have a meaning that directly reflects the prototypical meaning of the preposition. However, it is still possible that some kind of link can be established tying the meaning of the preposition in the complement construction to an aspect, or a figurative extension, of the prototypical or ideal meaning of the preposition.

As a preliminary to introducing later work, it is worth noting a remark on at and on by Knud Schibsbye. In his grammar he compares at and on and also discusses work at and work on as follows:

The value of point attached to at against that of ‘expanse’, ‘extent’ attached to on, which latter used figuratively leads to associations such as duration and/or accompanying circumstances is to be seen in cases such as: he would stay a few days at the farm / he works on the farm / / I looked at the hand / It was a new and fierce pain to look on all this (: ‘the well-loved room, pictures, rugs ...’) — Work at simply expresses the direction of the activity: he is working at a new invention / Alban worked at Chinese; work on includes duration: he is working on a poem // She was darning on a sock. (Schibsbye 1970: 327)

Schibsbye’s comments are a useful start, but they still leave room for a more fully worked out account.

To survey the ideal meanings of the two prepositions, Herskovits (1986) and Tyler and Evans (2003) are referred to here. Their accounts are limited to constructions of the type “at/on NP” and they do not discuss sentential complements. Here is Herskovits’s account of the ideal meaning of at:

The use types of at center around one ideal meaning:

at: for a point to coincide with another.

Because the first point corresponds to the Figure and the second to the Ground, the first is typically mapped onto an object and the second onto a fixed earth location. Most commonly, an expression with at asserts that those two points, each specified by a different description, overlap in space, and the various use types, listed below, can be seen as variations on this theme. (Herskovits 1986: 128)

Her use types include “Spatial entity at location,” as in Paul is at Yosemite, “Person at Institution,” as in Professor Jones is at Berkeley, “Person using artifact,” as in There is a student at the blackboard, and “Spatial entity at landmark in highlighted medium,” as in The bus is at the 3rd stop (Herskovits 1986: 128–136; the illustrations from the same source).

Tyler and Evans offer the brief comment that “at indicates co-location between a TR and a LM, in which the location is conceptualized as a point” (2003: 178) (For present purposes, “TR,” for “Trajector,” may be taken to correspond to “Figure” and “LM,” for “Landmark,” to “Ground.”)

Here is Herskovits’s analysis of the ideal meaning of on:

on: for a geometrical construct X to be contiguous with a line or surface Y; if Y is the surface of an object OY, and X is the space occupied by another object OX, for OY to support OX. (Herskovits 1986: 140)

Her use types include “Spatial entity supported by physical object,” as in the man on the chair, “Accident/object part of physical object,” as in the crack on the wall, “Physical object contiguous with another,” as in the lock on his forehead, “Spatial entity located on geographical location,” as in the players on the football field, and “Physical object contiguous with edge of geographical area,” as in a room on the patio (Herskovits 1986: 140–148; the illustrations from the same source). As for Tyler and Evans, they make the brief comment that “on mediates a relation of contact between a TR and a LM, which is a two-dimensional planar surface” (2003: 179).

The ideal meaning of at thus involves one point coinciding with another. In the case of the at ‑ing pattern, the Ground, using Herskovits’s terminology, corresponds to the gerundial complement, and the location denoted by the complement is then conceptualized as a point in the case of the at + NP construction. The action or activity named by the lower clause thus denotes a circumscribed task or point of activity in which the referent of the subject engages. To consider the example of at ‑ing cited above from Collins COBUILD, She works hard at keeping herself fit, a paraphrase for the construction work at Verb-ing might be ‘engage in (the task of) Verb-ing’.     

By contrast, in the case of the on ‑ing pattern, the Ground is conceptualized as a “two-dimensional planar surface.” This allows for a broader interpretation of the gerundial clause, with the gerundial clause conceptualized as a broader range of actions or a broader field of activity, and as a goal to be accomplished. There is therefore less focus on engaging in the circumscribed and specific action or activity expressed by the lower clause, and instead there is a suggestion or a possibility of ancillary activity that may lead to the goal named by the predicate of the lower clause. The example from Collins COBUILD also allows the preposition at to be replaced with the preposition on, as in She works hard on keeping herself fit, and a paraphrase for the construction work on Verb-ing might be ‘work towards (the goal of) Verb-ing’.     

The predicate keep oneself fit is atelic, without a natural culmination or end point. It is worth considering sentences with other atelic lower predicates and with different types of higher subjects, involving different scenarios. The sentences in (8a-b) are authentic, those in (9a-b) are invented, with the preposition changed:          

The idea of improvement, associated with the on pattern in Collins COBUILD (see above), seems less salient with the predicates collecting insect food and levelling the holes in (8a-b) than in the case of the predicate keep oneself fit. By contrast, the authentic at ‑ing complements in (8a-b) are in line with the specific and circumscribed nature of the scenarios denoted by the lower predicates.

Next, consider (10a-b) and (11a-b):

The lower predicates in the authentic examples (10a-b) – getting him released on bail and getting the ceremony started – are telic, with natural end points. Both variants are possible in these cases, but in (10a-b) there is again, in accordance with the broader interpretation of the on ‑ing pattern, more focus on the goals and on the notion of working towards those goals, while in the invented counterparts (11a-b) the focus is on the specific engagement that the referents are involving themselves in. The emphasis on goals in (10a-b) then also suggests that there is a more realistic prospect of accomplishing the goals in the case of (10a-b) than in the case of (11a-b).

Next consider the authentic sentence in (12) and its invented counterpart in (13):

In (12) on again conveys the idea of working towards a goal or taking steps with the objective of finding someone an apartment. For its part, at, with its point-like associations, suggests more circumscribed engagement in the task of finding someone an apartment.

There are also examples where one of the variants is considerably more likely than the other. Consider example (14):

Sentence (14) highlights engagement in the circumscribed task of practicing law, and an on variant, with its suggestion of working toward a goal, would seem unlikely here.

Consider also these examples with on ‑ing, which consist of pieces of dialogue:

The examples in (15) and (16) suggest that the on ‑ing construction is more flexible semantically than the at ‑ing construction. The lower predicates here are be bored and be cold, and work at being bored and work at being cold would convey the idea of ‘engage in the task of being bored/cold’ (or ‘engage in the task of becoming more bored/colder’). However, the on ‑ing complements found in the sentences do not convey the “normal” meaning of ‘work towards a goal’ either. Instead the meaning seems to be ‘take action in reaction to being bored/cold’ or ‘make an effort to remedy (the state of) being bored/cold’. This might be called a reactive meaning, with this meaning playfully suggesting sexual activity (as a remedy) in these examples. A reactive meaning does not seem conceivable with the at ‑ing pattern. It may be speculated that the possibility of a reactive meaning for on ‑ing may be related to the broader construal of the action or activity denoted by the gerundial complement in this case. A reactive meaning is rare in the data, but the flexibility of the on ‑ing pattern may still play some role in the rising frequency of this variant.

4. Quantifying the differences between the two constructions

As was observed in the previous section, based on authentic corpus data one can suggest differences between the work at -ing and work on -ing patterns. Sometimes such differences may be fairly slight, and it is by no means uncommon to find instances where the uses of the two patterns are very similar, as in the earlier cited examples (5a-b) and (6a-b). In these examples, the lower-level verbs expressing the actions engaged in by the subjects (making a success of the marriage, rebuilding our relationship, sharpening her intuitive skills, making your relationship rock solid) are very much the same, suggesting that the choice of the preposition makes no great difference as regards the meaning of the patterns. Indeed, attempting to examine some of the qualitative observations made in the previous section in quantitative terms is a challenging task, but some general points can nevertheless be made.

It was already observed in Section 2 that the work at -ing construction was found earlier in the COHA data. It may be worth noting that usually the 19^th century tokens described a specific, concrete task that was being engaged in, as in examples (17a-b):

Instances where the lower-level verb could be regarded as more broadly involving the notion of improvement or development, and where the activity is rather a long-term process towards a more or less clearly defined goal, only begin to appear in the 20^th century data. It does not appear from the COHA data that earliest references to efforts with a goal to improve something would have been made exclusively with the on ‑ing rather than the at ‑ing pattern. Instead, the broader sense was used with both patterns. It is nevertheless possible to see some differences between the two patterns as regards the occurrences of different verbs in the complements. While tokens of lower-level verbs such as building, developing, improving, and planning can be found with both the at ‑ing and on ‑ing constructions in the COHA data, it is interesting to observe that the data contains only one instance of work at developing but altogether 15 instances of work on developing, which perhaps reflects some preference for the use of the work on -ing pattern in cases expressing long-term development or improvement.

As for other lower-level verbs, the most prominent difference in terms of frequencies of occurrence that can be noted is the verb being. The verb is found in the data with both complementation patterns, as seen in (18a-b):

Some instances of work at/on being may not necessarily imply effort with a particular goal of improving or getting better at e.g. being friendly or annoying. Instead, it is possible that the people referred to in the examples already have these qualities, and that maintaining and adhering to these characteristics requires effort. With no long-term process being implied, it is perhaps indicative of the different general shades of meanings of the at ‑ing and on ‑ing patterns here that being follows work at considerably more often than it follows work on: the COHA data contains altogether 32 instances of work at being, while work on being is found only four times. This would support the observation made earlier of the at ‑ing pattern being associated with more specific, circumscribed, and momentary actions and processes. Although one might then assume that the verb becoming might, in contrast, be more common with work on ‑ing, work at becoming and work on becoming are about equally common in COHA (with only five and four tokens, respectively). Here it is likely that the preference for the on ‑ing pattern as a reflection of its closer association with broader, long-term efforts is not very prominent because the same idea is more idiomatically expressed with the to-infinitive (i.e. work to become).

Many tokens in the data also include the adverb hard modifying the matrix verb, and occurrences of elements of this type could be examined to see whether they are at all indicative of underlying semantic differences between the two constructions. In COHA, different types of adverbials (e.g. of duration or manner) are found with both work at ‑ing and work on -ing, and it is mainly the adverb hard that shows notable differences between its use with the two constructions. In the data from the period of 1990–2009 in COHA, hard was used 40 times with work at ‑ing, whereas it was found only five times with work on ‑ing. Considering that the total numbers of tokens of the two patterns in this period were 83 and 123, respectively, the more frequent use of hard with work at ‑ing may indeed be reflective of how the meanings of the two constructions are perceived. While the use of the adverb hard is by no means impossible with work on ‑ing, it is perhaps logical to assume that observations on the intensity of efforts are generally more likely made on single actions than on actions or projects that take more time.

In addition to examining the lower-level verbs and adverbs used with the work at ‑ing and work on ‑ing patterns, manual inspection of the corpus data prompted a look into whether the tenses and aspects of the matrix verb work themselves show differences between the two complement patterns. Tables 2 and 3 below include the numbers of instances of different forms of work when followed by at ‑ing and on ‑ing in the BNC and the 2000–09 data of COHA, respectively:

Based on the findings in Tables 2 and 3, it appears that the aspectual uses of the matrix verb work can be regarded as distinguishing between the work at ‑ing and work on ‑ing constructions, at least to some extent. While both constructions are found with simple (or non-progressive) forms, tokens of work on -ing have the matrix verb occurring proportionally more often in the progressive. In the COHA data, the division is clear, with only one instance found of the progressive use of work followed by an at ‑ing complement. A simple Chi-Square test of the total numbers of the simple and progressive forms of the matrix verbs shows statistical significance for the division in this respect in both British and American English: with the BNC data, the Chi-Square is 7.23 and the result is very significant (p < 0.01), and with the COHA data, the Chi-Square is 22.21 and the result is extremely significant (p < 0.0001).

These findings provide further support for the observations made earlier. The use of the progressive forms (such as is/was working, has/have/had been working, and auxiliary + be/have been working), as observed e.g. by Huddleston (1988: 74) is associated with situations in progress, whereas the simple or non-progressive constructions (e.g., worked) “can be used for both static and dynamic situations”, but the situation is nevertheless “presented in its totality”. Furthermore, in their analysis of the occurrence of simple and progressive forms of different verbs, Biber et al. (1999: 473) note that “verbs that rarely occur in the progressive fall into two main groups: (a) those that refer to an action that is immediate, and (b) those that refer to a state that is not normally a continuing process.” They also observe a connection between the duration of the action and the low frequency of the use of the verbs in the progressive, saying that with dynamic verbs such as shut, smash, and throw, “[b]ecause these actions have virtually no duration, such verbs rarely occur in the progressive” (Biber et al. 1999: 474). While it is clear that the verb work is obviously a dynamic verb when it is followed by either an at ‑ing or on ‑ing complement, one could suggest that the fact that work occurs more frequently in the progressive when followed by on ‑ing reflects a semantic difference between the two constructions. Compared to work at ‑ing, work on ‑ing could be seen as being more closely associated with the idea of improvement and with actions that are not momentary, but which require at least some time to complete. This observation in turn is in line with the ideas proposed earlier that the senses of the two constructions are influenced by the fundamental meanings of the two prepositions.

5. Concluding remarks

There are definite challenges in distinguishing between the constructions work at ‑ing and work on ‑ing even with the help of corpus data. It is perhaps not surprising that dictionaries do not clearly differentiate between the two, while some signs suggesting a difference in use are visible. One reason for this may be the fact that the major changes in the overall frequencies of the two patterns are still relatively recent. In addition, there appears to be some overlap in the uses of the two constructions. However, as has been observed in the qualitative and quantitative analyses of the corpus data, some lines of distinction can be detected as well, and the findings lend support to Bolinger’s view of a difference in form coinciding with a difference in meaning. It is worth considering that it is, in fact, the fundamental meanings of the prepositions at and on which are reflected in the uses of work at ‑ing and work on ‑ing. Overall, work at ‑ing seems to be preferred when referring to actions which are more specific and circumscribed, and the difference between the two constructions in terms of duration of the process is reflected in the occurrences of simple and progressive forms of the matrix verbs. Further, more in-depth examination is nevertheless warranted on the matter, which serves as a case study on the nuances that ultimately can be observed in the use of alternative constructions. Some possibilities of examining the issue in more detail could be found, for example, in the study of the two constructions in different text types as well as in diachronic study of the aspectual uses of the matrix verbs.

Sources

BNC = The British National Corpus: BNCweb CQP Edition. Developed by Sebastian Hoffmann and Stefan Evert. Available at http://www.natcorp.ox.ac.uk

COHA = The Corpus of Historical American English: 400 million words, 1810–2009. 2010–. Compiled by Mark Davies. Available at http://corpus.byu.edu/coha/

SOAP = The Corpus of American Soap Operas: 100 million words, 2001–2012. Compiled by Mark Davies. Available at http://corpus.byu.edu/soap/

References

Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad & Edward Finegan. 1999. Longman Grammar of Spoken and Written English. Harlow: Pearson Education Limited.

Bolinger, Dwight. 1968. “Entailment and the meaning of structures”. Glossa 2: 119–127.

Chomsky, Noam. 1985. Knowledge of Language: Nature, Origin, and Use. New York: Prager.

Herskovits, Annette. 1986. Language and Spatial Cognition. An Interdisciplinary Study of the Prepositions in English. Cambridge: Cambridge University Press.

Huddleston, Rodney. 1988. English Grammar: An Outline. Cambridge: Cambridge University Press.

Rohdenburg, Günter. 2006. “The role of functional constraints in the evolution of the English complementation system”. Syntax, Style and Grammatical Norms: English from 1500–2000, ed. by Christiane Dalton-Puffer, Dieter Kastovsky, Nicholas Ritt & Herbert Schendl, 143–166. Bern: Peter Lang.

Rundell, Michael, ed. 2002. Macmillan English Dictionary for Advanced Learners. Oxford: Macmillan Publishers.

Schibsbye, Knud. 1970. A Modern English Grammar with an Appendix on Semantically Related Prepositions. London: Oxford University Press.

Sinclair, John, ed. 1987. Collins COBUILD English Language Dictionary. London: HarperCollins Publishers.

Summers, Della et al. eds. 2003. Longman Dictionary of Contemporary English. Fourth edition. Harlow: Longman.

The Oxford English Dictionary (OED Online). 2015. Oxford: Oxford University Press. http://www.oed.com

Tyler, Andrea & Vyvyan Evans. 2003. The Semantics of English Prepositions: Spatial Scenes, Cognition and the Experiential Basis of Meaning. New York & Cambridge: Cambridge University Press.

Vosberg, Uwe. 2009. “Non-finite complements”. One Language, Two Grammars? Differences between British and American English, ed. by Günter Rohdenburg & Julia Schlüter, 212–227. Cambridge: Cambridge University Press.