Tuesday, 31 October, 2000
�Languages whose speakers conform to a minority default pattern present a major challenge to connectionist accounts of inflectional morphology.� Discuss.
Evaluate the current status of the notion �rule� as a representational entity.
The nature of underlying mental representations is revealed in children�s errors of the English past tense. Discuss.
How has connectionism contributed to the understanding of how inflectional morphology is acquired and processed?
Another influential early connectionist model was a net trained by Rumelhart and McClelland (1986) to predict the past tense of English verbs. The task is interesting because although most of the verbs in English (the regular verbs) form the past tense by adding the suffix �ed�, many of the most frequently verbs are irregular (is / was, come / came, go / went). The net was first trained on a set containing a large number of irregular verbs, and later on a set of 460 verbs containing mostly regulars. The net learned the past tenses of the 460 verbs in about 200 rounds of training, and it generalized fairly well to verbs not in the training set. It even showed a good appreciation of "regularities" to be found among the irregular verbs (send / sent, build / built; blow / blew, fly / flew). During learning, as the system was exposed to the training set containing more regular verbs, it had a tendency to overregularize, i.e. to combine both irregular and regular forms: (break / broked, instead of break / broke). This was corrected with more training. It is interesting to note that children are known to exhibit the same tendency to overregularize during language learning. However, there is hot debate over whether Rumelhart and McClelland�s is a good model of how humans actually learn and process verb endings. For example, (Pinker & Prince 1988) point out that the model does a poor job of generalizing to some novel regular verbs. They believe that this is a sign of a basic failing in connectionist models. Nets may be good at making associations and matching patterns, but they have fundamental limitations in mastering general rules such as the formation of the regular past tense. These complaints raise an important issue for connectionist modelers, namely whether nets can generalize properly to master cognitive tasks involving rules. Despite Pinker and Prince�s objections, many connectionists believe that generalization of the right kind is still possible (Niklasson and van Gelder, 1994).
grammar Not to be confused with socially correct usage. In order to handle novel sentences, we not only need to access the words stored in our brains but also the patterns of sentences possible in a particular language. These patterns describe not just patterns of words but also patterns of patterns. There are three aspects of grammar: morphology (word forms and endings), syntax (from the Greek "to arrange together" � the ordering of words into clauses and sentences), and phonology (speech sounds and their arrangements). A complete collection of rules is called the mental grammar of the language, or grammar for short.
grammatical morphemes The words (a few dozen in English) that refer to relations between content words. They are unlike content words, which refer to concepts of things in the world. They include words that express relative location (above, below, in, on, at, by, next to), relative direction (to, from, through, left, right, up, down), relative time (before, after, while, and the various indicators of tense), and relative number (many, few, some, the _s of plurality). The articles express a presumed familiarity or unfamiliarity (the for things the speaker thinks the hearer will recognize, a or an for things the speaker thinks the hearer won't recognize) in a manner somewhat like pronouns. Others express relative possibility (can, may, might), relative contingency (unless, although, until, because), possession (of, the possessive version of _s, have), agency (by), purpose (for), necessity (must, have to), obligation (should, ought to), existence (be), nonexistence (no, none, not, un_), and so forth. These are called " closed-class words" because our ways of expressing relationships are so resistant to augmentation, whereas you can always create new nouns or verbs.
inflections The inflectional system of English alters a noun when it refers to a multiplicity ("The boy ate three cookie." Is that correct English?) and alters a verb when it refers to past time ,OK?). Late learners of English may fail to realize that anything is "wrong" with these incorrect sentences, as such long-range dependencies are redundant information that helps out in noisy environments when some words are imperfectly heard and must be guessed.
word order A simple convention that aids in identifying roles, such as the subject-verb-object order (SVO) of most declarative sentences in English ("The dog bit the boy") or the SOV of Japanese. At least in English, the who-what-where-when-why-how questions deviate from basic word order: "What did John give to Betty?" is the usual convention (except on quiz shows in which questions mimic the basic word order and use emphasis instead: "John gave what to Betty?"). Some languages such as Latin lack a systematic word order, instead using characteristic inflections or even separate words (as when English uses "he" for a subject and "him" for an object, although both have singular, masculine, third-person referents) to help disambiguate the sentence
Morphology is concerned with the units, called morphemes, that carry meaning in a language. These may be word roots (as the English cran-, in cranberry) or individual words (in English, bird, ask, charm); word endings (as the English -s for plural: birds, -ed for past tense: asked, -ing for present participle: charming); prefixes and suffixes (e.g., English pre- , as in preadmission, or -ness, in openness); and even internal alterations indicating such grammatical categories as tense (English sing-sang), number (English mouse-mice), or case.
These are units of meaning, either a stem or affix (prefix, suffix or infix � found in languages such as Swahili, but not English). They can be either derivational or inflectional. Inflectional morphemes change the grammatical status of the word, such as the suffix �-ed� which indicates the past tense, whereas derivational morphemes, such as �un-�, change the meaning of the word. Words are comprised of at least one morpheme.
Regularisation is the means infants use to acquire the set of complex rules which determine grammar. Just from hearing instances of adults speaking and using these rules, they are able to differentiate and abstract to the rules governing the production of grammatical sentences. However, in cases such as �bring-brought�, the application of regular rules which apply to regular verbs, �talk-talked� or even irregular verbs �run-ran�, would be inappropriate, i.e. over-regularisation.
After almost two years, children start to combine words, usually in pairs, such as �Adam fall� and �allgone ball� to describe events and actions. Brown also tracked a sequential development of the use of morphemes during the third year, such as inflectional affixes.
Marcus et al�s studies on over-regularisation (applying regularisation rules such as �run-ran� in the past tense to �bring� to get �brung�) indicate a conflict in the sequence of their semantic and syntactic developments, as they initially learn the correct form, then learn the regularisation rule and then apply it too widely.
u-learning � for instance, infants first learn all verbs by rote as a series of exceptions. it is only later that they abstract to the rules governing the use of verbs, i.e. the regular onesand the morphology governing their inflection(???), e.g. the tenses, conjugation. indeed they overregularise, understandably but mistakenly extending the use of such rules to irregular verbs that they previously used correctly � it is only later still that language often consists of general rules with many exceptions
what alternatives to a system of rules are there?
langauge as a huge finite state machine � we learn all the possible sentences in the world and then select the appropriate one � ridiculous because of combinatorial explosion
word chains � slightly abstracted, shows possible phrases as trees of categories, say, with sentences being comprised of chains of such phrases. again, not possible, because of the totally modular way in which we can combine phrases, e.g. recursively, and with long-distance dependencies � again, the word chains would become impossibly long
so we are left with a system of rules that somehow allow us to determine and generate legitimate use of language.
plunkett talks about 2 different types of rules/device, a LAD and a language generation device???
Why are we particularly concerned with inflectional as opposed to derivational morphology?