The spelling of the regular past tense in English: implications for lexical spelling and dual process models

Vivian Cook  Writings Home  Chomsky's Optimal Spelling

In G. Bergh, J. Herriman & M. Mabrg (eds.) AIMO: An International Master of Syntax 2004.
This paper uses the phonetic font SILDOULOS IPA which can be downloaded from the Summer Institute of Linguistics (SIL). Without it some of the symbols in the phonetic transcripts may appear odd (on the other hand they all represent normal sounds of English and so you probably won't miss much). Note: /..../ encloses phonetic script, <....> orthographic symbols, i.e. letters and punctuation. 

During the 1990s the English past tense figured prominently in debates about language acquisition and processing: did language acquisition mean acquiring general rules that could be applied to many items or did it mean forming associations between many idiosyncratic items? The forms of the English past tense constituted a test-case since both possibilities seemed to exist within the same area: on the one hand rules for forming regular forms, on the other large numbers of irregular exceptions that had to be learnt one at a time. Connectionists attempted to show that all the past tense forms whether regular or irregular could be learnt without rules (Rumelhart & McLelland, 1986). Others showed that rules were necessary to explain how people assigned past tenses to novel verbs, 'grok' becoming 'grokked'. Pinker and his associates argued for the compromise position that both rule-learning and instance learning were necessary (e.g. Pinker & Prasada, 1993).

At the same time the English past tense also played a crucial role in theories of spelling. Here the emphasis was on the extent to which English spelling went beyond sound-to-letter correspondence rules. While spoken English has three ways of pronouncing the past tense 'ed', written English uses a single <ed> spelling. Hence Noam Chomsky argued that English spelling reflected underlying lexical rather than phonological representation (Chomsky, 1972); it connects to the morphemes in the mental lexicon, only indirectly to the phonemes of speech. This will be called 'lexical spelling' here, following Carol Chomsky's use (Chomsky, 1970).

For both these areas of debate the crucial point was the difference between regular and irregular forms: if everything were regular, rules would reign supreme; if everything were irregular, instances would reign. By bringing together different evidence presented in Cook (2004), the present paper argues that both debates ignored the features of the past tense in the English writing system: to what extent does English spelling actually support the two claims which both these approaches make use of?

The forms of the English past tense in speech and writing

The starting point is the comparison of the written and spoken forms. In modern spoken English the past tense, inflection 'ed' has three regular forms depending on the final phoneme of the verb stem, as described in any English grammar, say Greenbaum (1996):

There are also several groups of irregular past tense forms. To take a sample:

Minor variations exist between British and US usage where contemporary US English seems to have 'dove' and 'fit' to British 'dived' and 'fitted'.

On the other hand modern written English has four regular rules:

In addition there are several groups of irregular past tense forms, including:

Figure 1 Rules for the forms of the English past tense inflection 



S-reg1. addition of /t/ /kIkt/

W-reg1. addition of <ed> 'waited'

S-reg2. addition of /d/ /peId/

W-reg2. addition of <d> after <e> 'saved'

S-reg3. addition of /id/ /weitId/

W-reg3. doubling <VcCC>+<ed> 'sobbed'


W-reg4. <Cy> → <Cied> 'cried'

S-irreg1. no change /hit/

W-irreg1. no change 'hit'

S-irreg2. change of vowel only /rQn/

W-irreg2. change of vowel only 'ran'

S-irreg3. change of whole word /went/

W-irreg3. change of whole word 'went'

S-irreg4. change of vowel plus
/d~t/ /dId/
/v/ → /ft/ /left/
/i:p/ → /ept/ /kept/, /slept/
/n/ →/u/ /stud/

W-irreg4. change of vowel and consonant
<d> 'did'
<ve> →<ft> 'left'
<ep> → <pt> 'kept', 'slept'
<n> 'stood'
<y> → <i> +<d> 'paid'


W-irreg5. <Vy> → <aid> 'paid'


W-irreg6. addition of <t> 'spoilt'


W-irreg7. change of vowel+<ght> 'bought'


The two systems are juxtaposed in Figure 1. Spoken English has three regular rules, S-reg1 to S-reg3, concerned with the three possible phonological forms of the past tense <ed>. Written English has four regular rules, W-reg1 to W-reg4, adding <ed> as a default, stopping <e> being doubled when the word stem already ends in <e>, doubling the final consonant to preserve the 'checked' preceding vowel a widely-used rule of English not just specific to the past tense (Venezky, 1999) and changing <y> to <i> following a consonant 'cried' again a rule with wide implications such as the plural 'countries' etc. Written English ignores the phonological distinctions in the three spoken rules but adds three regular rules to deal with other spelling correspondences.

The first three groups of irregular verbs, are essentially identical in speech and writing: no change 'hit', change of vowel only 'read', change of whole word 'went'. The group of verbs that change both vowel and consonant, those with /d/ <d> 'did', /v/ <ve> 'left', /i:p/ <ep> 'slept' and /n/ <n> 'stood' seem equivalent in S-irreg4 and W-irreg4. However there are additional irregular forms in writing, namely the change of <y> to <i> +<d> 'paid' (an exception to the regular rule W-reg4 <Cy> to <Cied> 'cried'). Their spoken counterparts /peId/ and /kraId/ are not, of course, irregular in speech but covered by S-reg3. An additional group of verbs maintain the vowel in writing with the addition of <t> 'learnt', though many of these also have <ed> variants 'learned'; since these are not covered by the regular written rules, they are irregular in writing. Finally some written verbs change the vowel and then add a <gh> followed by <t> 'taught'. In the spoken language these are members of the large group of vowel change plus consonant since <gh> has no spoken correspondence.

There are then differences in the definition of regular and irregular in speech and writing, both in what constitutes a regular form and in the groups of irregular forms.

Learners' development of 'ed' spellings

The rule that the regular spelling for the past tense is <ed> W-reg1 has been taken as the classical example of lexical spelling in English: the written form <ed> corresponds, not to a single pair of sounds /I d/, but to a morpheme for past tense that can be said in three ways /t~d~Id/. This is central to the N. Chomsky/M. Halle/C. Chomsky argument that English spelling is lexical rather than phonological representation, for example Chomsky (1972), and treated as a paradigm example of linguistic rules by researchers into rule-based learning.

If this is indeed the case, native children would find it a boon to have a constant morphemic spelling in <ed> despite vagaries of pronunciation. So they would be quick to learn to treat the different spoken forms as a single morpheme and would make few mistakes with its spelling. The main regular written rule W-reg1 is then a test-case for both the lexical spelling theory and for rule-based research. If children had difficulties with this rule, if they did not treat the three spoken forms as having a single written form, then the evidence used in the two debates would be less convincing.

An experiment by Bryant, Nunes and Bindman (1997) investigated English children's spelling of past tenses between the ages of 6 and 8. This tested regular and irregular past tense forms and non-verbs with the same phonological endings, such as 'cold' and 'soft'. The average scores out of ten test items are shown in Figure 2.

Clearly the six-year-olds are not very good at spelling the past tense, getting only 2.28 correct out of 10 for the regular <ed>, with slightly better results for the irregular verbs with 5.98 and non-Verbs 6.04. The children's ability to spell the regular and irregular verbs increases steadily between 6 and 8 but the pattern stays the same for the older children. By the age of 8 they are achieving 5.68 for regular verbs, 7.9 for irregular verbs and 8.25 for non-verbs.

In other words, the regular verbs have the worst scores out of the three categories at each age. Far from helping the child to learn the English past tense, the common <ed> spelling hinders acquisition compared to the irregular verbs. Indeed, at the age of 8, children still only get 57% of regular past tenses correct. Treiman (1993) found grammatical inflections were one of the worst spelling problems for children, not just the past tense 'ed' but also the third person 's' and the plural 's', all of which share morpheme-based spelling.

The common written representation for the different form of <ed> might, however, facilitate adult L2 learners of English. I tested 65 adult L2 students of English with diverse first languages in language schools in Cambridge and compared them with the L1 children tested by Bryant et al (1997) seen above, using a similar test of regular verbs, irregular verbs and non-verbs. Results are shown in Figure 3.

The scores of the adult L2 learners are much higher than those of the L1 children, ranging from 8.82 for regular <ed>, 8.43 for irregular verbs and 8.65 for non-verbs. As a visual comparison of L1 and L2 learners in figure 3 shows, this is unlike the pattern for the children: all three types have about the same scores. So perhaps at least L2 learners benefit from the morphemic spelling. One reason might be that this feature of English has been directly taught to them, another that it is something which adults can exploit more readily than children. Or perhaps they are more literate than the children and this has altered their approach to spelling.

Research into the development of the past tense forms therefore supports neither the lexical spelling nor rule-based learning. Children do not start by seeing rules but have to acquire them over time. If anything, W-reg1 is a hindrance to their development another quirky feature of English spelling to master rather than an aid.

Historical development of 'ed' spellings

Another relevant area is how the spelling of past tense developed in the history of the English writing system. If the regular rule were such an asset to processing and to acquisition, one might argue that it should have existed ever since English started to be written. Alternatively, if it came into being at a particular historical point, one might look for its sudden appearance in writing and its immediate universal adoption. The question is when the uniform spelling of the past tense morpheme incorporated in rule W-reg1 manifested itself in the development of the English spelling system.

The evidence comes from equivalent texts from different periods. The sources were original editions of plays, starting from 1592 The Spanish Tragedie, attributed to Thomas Kyd, including a play about every twenty years up to Sheridan's School for Scandal in 1780, with a gap around 1740 filled by the dialogue from an edited novel, Fielding's Joseph Andrews 1742. All the examples of regular past tense were included up to about one hundred, the lowest number being 51 for 1667, Dryden's The Indian Emperour. More information is given in Cook (2004).

The historical development of the spelling of the regular past tense has three main features:

i) dominance, then decline, of <d>. From 1613 to 1760 the <'ed> spelling was most frequent, whether 'chargd' (1613) or 'preachd' (1760). <d> corresponded both to /d/ 'liud' (1613), 'forgd' (1633) and 'fird' (1760), and to /t/ 'promiߒd' (1613), 'hopd' (1633) and 'escapd' (1760).

ii) rise of <ed>. While the <ed> spelling started at 44% in 1592 as in 'loued', it was still only 20% in 1760. From 1780 it became effectively the only regular past tense form at 97%. In most early examples <ed> corresponded to the usual /id/ after /t/ or /d/, as in 'parted' (1633) or 'descended' (1742); in some cases to the full stressed form /id/ after other consonants, as in 'strowed' (1592) or 'walked' (1633).

iii) fall of <t>, <t> and <d>. Spellings with <t> started at a combined 29% in 1613 but rapidly declined, apart from a rise for <t> in 1742. Starting at 13% in 1592, <d> disppears rapidly, just as <de> is non-existent after 1592.

So at the earliest period studied <d> is used for past tenses regardless of whether it corresponds to /t/ 'lookd' (1633) or to /d/ 'hangd' (1633). The <ed> spelling corresponds to a spoken /I d/ till 1780 when it becomes the invariant lexical spelling for regular 'ed', as seen in Figure 4. Until 1780 there were at least two alternatives for the regular past tense 'ed'.

The consistent spelling of past tense as <ed> is then relatively recent in historical terms. As in children's development, there was a phase in the history of the English language when the different spoken forms of the past tense were reflected in different written forms, with a gradual development of <ed> as the standard regular form over a long period of time. Far from being an integral part of the written language since time immemorial the rule W-reg1 was fixed only in the eighteenth century. Hence it seems a late and superficial addition to the writing system, no more important than the development of the apostrophe in the seventeenth century and the curious division of <its> from <its> in the eighteenth. The <ed> regular spelling is not integral to English (unless one dismisses written English pre-1780) and came into being, not as a sudden invention of a rule, but as a slow evolution. Such development hardly gives it mythic status as evidence for lexical spelling or for the existence of rules.

Implications for the lexical representation view of spelling

The lexical spelling theory fails to find support from children's development since they find the morphemic spelling of little help; at 8, the score for regular <ed> is still only 5.68 out of 10. Treiman (1993) indeed found this deficiency extended to other morpheme spellings, not just the past tense 'ed' but also the third person 's' and the plural 's'.

Nor does the history of English support lexical spelling. On the one hand English managed for centuries to do without morphemic spelling for the past tense, using a fairly phonological spelling with <d>, <t>, and so on. On the other, the single morphemic spelling did not come into existence as a whole at one particular point but gradually coalesced as other spellings fell out of use. In addition Yule (1978) found that under 3% of the 6000 words set for children in the Victorial Education Department conformed to lexical spelling.

Implications for the dual process view of acquisition

The dual process model of language acquisition tries to balance rule and instance learning; Prasada and Pinker (1993) claim to have shown that a need for a 'hybrid' model where some aspects of the past tense are leant by rule, some by instance. However, as we have seen, their experiments do not deal with the rules or instances of written past tense. If they had been testing purely spoken stimuli which the subjects never saw written down, this would not be relevant. However their method involved giving a written sentence with a novel verb which subjects have to rate on a seven-point scale, in other words a test of putative spoken forms via a written form. Their hypotheses should then have been framed in terms of the rules of the written past tense form, in which the <ed> morphemic rule subsumes the three realisation rules of the spoken language. The groups of irregular forms should have been based on the irregularities of the written form where <paid> is irregular , not of the spoken where /paId/ is regular. It is not surprising they find the need for a 'hybrid' when their experiment relies upon a hybrid of spoken forms tested via writing.

This might not perhaps matter in a 'shallow' writing system like Finnish or Italian where the written language corresponds closely to the spoken form (Katz & Frost, 1992). English, however, is far from shallow and has multiple correspondences in both directions: the letter <a> corresponds to eleven phonemes, the phoneme /Q/ to at least twelve letters. English needs complex rules to bridge the spoken and written languages, as described in work such as Albrow (1972), Venezky (199) and indeed Chomsky (1972). Prasada and Pinker (1993), blithely ignore any such rules in their creation of non-verbs. Take their central example of 'ploamph', which they say has 'unusual sounds': why should the  correspondence for /f/ be <ph>, which occurs fairly uncommonly in word-final position ('Ralph', 'autograph'), compared to <f>? In other words 'ploamf' would be a far more typical written form, though still equally a non-verb; they have compounded the 'unusual sounds' by creating unnecessarily unusual spelling. The structure of their non-verb stimuli is skewed throughout by the use of extremely rare spellings: 'skring' rather than 'scring' the only word with initial <skr> in the 100 million British National Corpus (BNC) is 'skrew', not found in the OED (1994) at all); 'smairph' the only word with final <-rph> in the BNC is 'biomorph'; 'smeerg' with <-erg>, only found word-finally in the BNC with proper names and words derived from 'berg', 'ploanth' end in '-anth', found only in the BNC in 'amaranth' 'coelacanth' and 'srikkanth' '. These form examples of what Albrow (1972) calls the exotic spelling system for English for handling new words rather than of the basic or Romance systems he describes. The testing of the spoken language forms is distorted by the imposition of unnecessary spelling deviancies1. There may indeed be relevant evidence for the dual 'hybrid' model from the written past tense but it cannot be deduced from Prasada and Pinker (1994).

General conclusion

The final conclusion is then that linguists and psycholinguists are paying insufficient attention to the well-known facts of English spelling when devising their theories and experiments. Their claims for lexical spelling and the dual process are not supported by past tense evidence when a proper account of the written past tense is used.


1. Indeed the authors' insensitivity to the writing system is evidenced by the presence of two classic spelling mistakes in their paper: 'the principle methodological challenge' (p.8) and 'necesary' (p.9).


Albrow, K.H. (1972). The English Writing System: Notes towards a Description. London: Longman.

Bryant, P.E., Nunes, T. & Bindman, M. (1997). Children's understanding of the connection between grammar and spelling. In B. Blachman (ed.). Linguistic underpinnings of reading. Hillsdale, NJ: Erlbaum, 219-240.

Chomsky, C. (1970). Reading, writing and phonology. Harvard Educational Review 40, 2, 287-309.

Chomsky, N. (1972). Phonology and reading. In Levin, H. (ed.) Basic Processes in Reading. Harper and Row, 3-18.

Cook, V.J. (2004). The English Writing System. London: Edward Arnold

Greenbaum, S. (1996). The Oxford English Grammar. Oxford: Oxford University Press

Katz, L. & Frost, R. (1992). Reading in different orthographies: the orthographic depth hypothesis. In R. Frost & L. Katz (eds.). Orthography, Phonology, Morphology and Meaning. Elsevier, 67-84

Oxford English Dictionary (OED) (1994). CD-ROM version 1.13. Oxford University Press.

Prasada, S. & Pinker, S. (1993). Generalisations of regular and irregular morphology. Language and Cognitive Processes, 8, 1-56

Rumelhart, D.E. & McLelland, J.L. (1986). One learning the past tenses of English verbs', in McLelland, J.L., Rumelhart, D.E. & the PDP Research Group, Parallel Distributed Processing: Volume 2 Psychological and Biological Models, Cambridge Ma; MIT Press.

Treiman, R. (1993). Beginning to Spell: A Study of First-Grade Children. Oxford: Oxford University Press.

Venezky, R.L. (1999). The American Way of Spelling. New York: Guilford Press.

Yule, V. (1978). Is there evidence for Chomsky's interpretation of English spelling? Spelling Processes Bulletin 18, 4, 10-12.