Joyce Tang Boyland's study of
'would+have+past-participle'

Handout for LING 319/519/SARS 319/519
Topics in Dravidian Linguistics:

GRAMMATICALIZATION

1. Introduction

   English would+have+past-participle' has been involved in a cluster of interesting phenomena:

   • Spelling errors: would of

   • Uncertainty of syntactic judgements:

     'Should-ve we acted earlier?
     Should-ve we gone in and wiped him [Saddam] out in 1991?' (Heard on radio interview, HS, 2003)

   • Others (see below)

   Does this mean something is being grammaticalized? JTB proposes that what is happening is morphologization (what H&T call a phase of grammaticalization )

   JTB presents quantitative data to show that have is being more 'tightly bound' to would:

   would [have PAST PARTICIPLE...]--> [ would have ] PAST PARTICIPLE...

   This seems to indicate that have is becoming affixed to would which is predictible because this is the most frequently occurring modal before have. The phonetic change (phonological reduction) and the morphosyntactic change are going together, and there is furthermore a cognitive process called chunking which together with the perceptual process of auditory grouping that are the plausible mechanisms for this.

2. Background. When a sentence like We should have taken him out in 1991 is made interrogative with a WH-Q like 'when' we can get:
   When should we have taken him out?

   But we can also get (non-standard)

   When should we of taken him out?

   This only occurs in writing, since the vowel of of (if there is one at all) is the same as the vowel of 'have' i.e. [@v] (where @ = schwa).

   This is common in American non-standard, but also now appearing in British. Result: loss of 'identity' of the AUX 'have' and 'merger' (?) with the modal.

   When the conditional is used, even more reduction:

   If I had run faster... ---> If I'd a run faster

   which may actually be a reduction of (nonstandard, but 'hypercorrect')
   If I would have run faster... (instead of If I had..)

   Americans interpret 'd'a as underlyingly would have but Brits interpret it as had have (Harris 1984). (Discussion about what underlies the schwa here by various sources). Whatever the case, these elements are losing identity as separate pieces, and merging.

   Note all the NICE phenomena: (NICE = Negation, Inversion, Code, and Emphasis). (Examples in 7)

   One striking inversion is the one I already noted (above):

   What should've you done?

   Various reactions to and attitudes about this in terms of grammaticality. (See pg. 3)

   Can also involve fronting of other stuff, e.g. all as in 'But it would've all availed me nothing' instead of 'But it would all have ..." (example from 1840).

   This context (would-have) is the only one where 'have' can be contracted. (Gives examples of would-ve vs. starred *Wood-ve . (Proper name Wood).

   Also, this contraction occurs sometimes with preterit, not past-participle:

   • You should've took precautions.
   • You've *took precautions.

   Variation in function: the de-categorialization of woulda, coulda, shoulda which seem to be used as Nouns.

   Another peculiarity: repeated haves (examples in 14). Perhaps functions of have are splitting, with one incorporated as 'counterfactual' (and attached to would ) and the other for past-marking.

3. Adverb Placement

   This study is based on advert placement and parentheticals from real usage corpora. The claim that grammaticalization is occurring must be based on data that show frequent usage. Since adverbs can occur "in any slot" we need data that show the choice of slotin normal usage, to show

   • which can be easily separated vs.
   • which tend to remain adjacent.

   		SLOT 1		SLOT 2
   SUBJ	would		have		PAST-PARTICIPLE

   Study concentrated on would+have which are more frequent (479) than should+have and could+have (348 combined).

   Question: are interruptions more frequent in Slot 1 or in Slot 2?

   • What constitutes an interruption?

   • Adverbials and parentheticals, but not and never were excluded because negation shows tight affinity with the modal, i.e. makes for 'knock-out' context giving high selection for Slot 1. Only Prose selections, because rhythm is important (in promoting grammaticalization).

   • Page 7 gives data sources, dating back to 1500, but most from 1960 on, lots of 1990's. Two charts (page 10) showing British and American usage. The higher up the y-axis, the greater proportion of interruptions in Slot 2. The higher the point, the more often interruptions occur after have meaning the more often would + have remain adjacent.

   • Data show that in EModE and 18th-19th C data, interruptions in Slot 2 are rare.

   • Speech data vs. Writing: earlier data are writing, so few examples. (I did a Google search, and no examples of written would+of etc. appear. Only the examples cited earlier woulda, shoulda, coulda as a nominal phrase.)

   • Later spoken corpora show as much as 63% Slot 2 interruptions, and American spoken corpora show as high as 80% at Slot 2.

   • In written corpora when interruptions at slot 2 occur, it is in informal genres (predictible!) that they occur.

     Summary: the more spontaneous the data, and the more American the data, the more preference for Slot 2 interruptions, in order to maintain the adjacency of would + have at the expense of have and the PPl.

4. Interpretation
   • Straightforward: Degree of affinity betw. modal and 'have' is increasing, in genres where least prescriptivism is active. Early step toward morphologization, which we only see in retrospect.

   • Maybe this has nothing to do with grammaticalization of the modal perfect? (Could be that adverbs are changing their positions with respect to all aux's.) However, whatever the cause of change, resulting change in usage patterns prepares the way for morphologization to occur, since

     'grammaticalization characteristically transforms non-obligatory patterns into entrenched obligatory patterns.' (Boyland 1998:11)

   • More objections: not enough data in earlier corpora. Maybe the data (modern) are all 'register-based' meaning it's not historical change but just not reflected in the older data. But some data do show a rise in American data, esp. in writing, esp. non-standard writing (culled from letters, etc.)

   • Data showing greatest deviations from conservative standard are found in least conservative genres. If morphologization is occurring, what is the cause? Possible mechanism from cognitive psychology, known as chunking.
     • First: memory encoding (1956)
     • Later: problem solving (1993)
     • Skill acquisition (1987)
     • Learning implicit knowledge (1990) eg. knowledge of structure that is unconscious but can be described as grammatical.

     What would it mean for chunking to be involved in grammaticalization?

     Three stages:

     1. Process of understanding a structure (in language): at first all confused and bewildering.

     2. Then, rules can be discerned, partially analyzable into steps or parts.

     3. Then complex structure resolidified into a Unitary Experience. Chunking is the process that moves speakers from 2nd stage to 3rd.
     Examples from learning to drive a car, especially a stick shift At first, confusing, jerky, unintegrated; then steps are discerned; then, process becomes a whole or CHUNK which some people describe as 'automatic.'
     • Characteristics of chunking make it a likely candidate for the bonding of would and have and for the stages of late grammaticalization i.e.

       'the routinization and compacting of one or more elements that co-occur frequently in a functional context.

     • Conditions are also similar: high frequency of occurrence is the key requirement for linking complex sequence to functional context.

     • Relative frequency of co-occurrence with each other is necessary for elements to become bonded.

     • Chunking is gradual learning for both children and adults.

     • Applied to would+have they are the most frequent, so should show most bonding;

     • Would and have are 'closed class' items, and occur with each other more than with any arbitrary PastParticiple, which is from an open set.

   • Another possible process: Perceptual Grouping. (Langacker 1994). Applies primarily in the listener whereas chunking applies in the speaker. Reasons:
     1. Natural speech packaged in prosodic units, which are not just ling. units but perceptual units. Prosody may not be coded, but prosody gives structure that is perceived by listener.

     2. Perceptual Grouping easily accepts hierarchical structure such as in language.

     3. Perceptually grouped sequence may be more easily chunked since it is already a unit.

        'Part of the work of chunking, that is finding what elements belong together, is done already if the elements belong to the same perceptual group.'

     4. Specifically: grammaticalizing forms typically become unstressed (phonological reduction.)

     5. Unstreessed pieces of sound tend to attach themselves to nearby more stressed pieces.

     6. The piece they attach to depends on

        • acoustic structure,
        • but also (b) expectation, which comes from experience.
        • In English, weak syllables most often attach to a preceding (not following) syllable.

        • It's unclear how to assign the weak syllable, as in would+have+ PAST-PARTICIPLE sequences, so when in doubt, assign it to the preceding syllable not the following.

        • Another good reason why have is more linked to the preceding would than following stuff.

5. Thus, Relation betwen prosody and other levels of ling structure, and between prosody and perceptual grouping, leads to (or allows) morphemes to excape their syntactic and lexical origin and get attached to other things (example of gonna where -na was once the infinitival to.

6. Conclusion

   1. Corpus data show increased affinity (closeness) between would and have esp. in least formal registers (spoken language.)

   2. This is a step toward morphologization (erasure of morpheme boundary).

   3. Cognitive processes of chunking and auditory grouping perform major roles in this mechanism.

      ---

      haroldfs@ccat.sas.upenn.edu
      last modified 3/15/05