Silvia Rădulescu

Trace:

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
aslin_newport [2015/11/22 23:30] – silviaaslin_newport [2016/02/13 11:03] (current) – silvia
Line 1: Line 1:
-====== Statistical Learning: From Acquiring +====== Statistical Learning: From Acquiring Specific Items to Forming General Rules ======
-Specific Items to Forming General Rules ======+
  
 ---- ----
Line 56: Line 55:
 the same rules. the same rules.
 \\ \\
-Some researchers have claimed that statistical learning and+**Some researchers have claimed that statistical learning and
 rule learning are two separate mechanisms, because statistical rule learning are two separate mechanisms, because statistical
 learning involves learning about elements that have been presented learning involves learning about elements that have been presented
 during exposure, whereas rule learning can be applied during exposure, whereas rule learning can be applied
-to novel elements and novel combinations (see Endress & +to novel elements and novel combinations** (see Endress & 
-Bonatti, 2007; Marcus, 2000). But why do learners sometimes+Bonatti, 2007; Marcus, 2000). **But why do learners sometimes
 keep track of the specific elements in the input they are keep track of the specific elements in the input they are
 exposed to and at other times learn a rule that extends beyond exposed to and at other times learn a rule that extends beyond
 the specifics of the input? An alternate hypothesis is that these the specifics of the input? An alternate hypothesis is that these
 two processes are in fact not distinct, but rather are different two processes are in fact not distinct, but rather are different
-outcomes of the same learning mechanism.+outcomes of the same learning mechanism.** 
 +\\ 
 +\\ 
 +//MyNote//:this is the most relevant question they ask in this study. 
 +\\
 \\ \\
 For example, some stimulus dimensions are naturally more For example, some stimulus dimensions are naturally more
-salient than others. If stimuli are encoded in terms of their+salient than others. **If stimuli are encoded in terms of their
 salient dimensions rather than their specific details, then learners salient dimensions rather than their specific details, then learners
 will appear to generalize a rule by applying it to all stimuli will appear to generalize a rule by applying it to all stimuli
-that exhibit the same pattern on these salient dimensions.+that exhibit the same pattern on these salient dimensions.** 
 +\\ 
 +\\ 
 +//MyNote//: what triggers encoding in terms of the salient dimensions that apply to all stimuli? 
 +\\
 \\ \\
 Although perceptual cues can serve as powerful constraints on Although perceptual cues can serve as powerful constraints on
Line 78: Line 85:
 are defined in the natural environment. are defined in the natural environment.
 \\ \\
-They acquire rules when patterns in the input indicate+**They acquire rules when patterns in the input indicate
 that several elements occur interchangeably in the same contexts, but acquire specific instances when the patterns that several elements occur interchangeably in the same contexts, but acquire specific instances when the patterns
-apply only to the individual elements. For example, Xu and+apply only to the individual elements.**  
 +\\ 
 +\\ 
 +//MyNote//: **CRUCIAL POINT**: what features of the input indicate that elements occur interchangeably? How much evidence is needed to this end for generalization to new elements to occur? 
 +\\ 
 +\\ 
 +For example, Xu and
 Tenenbaum (2007) have shown that if children hear the word Tenenbaum (2007) have shown that if children hear the word
 “glim” applied to three different dogs, they will infer that “glim” applied to three different dogs, they will infer that
Line 114: Line 127:
 ended in four different syllables, only the AAB rule was ended in four different syllables, only the AAB rule was
 reliable. reliable.
 +\\
 +\\
 +//MyNote//: **QUESTION**: 
 +\\
 +Consider the set of 4 strings: //leledi, wiwije, jijili, dedewe//
 +\\
 +The following rules are equally reliable for all strings:
 +\\
 +1. AAB
 +\\
 +2. starts with 2x //le, wi, ji or de//
 +\\
 +3. ends in //di, je, li, we//
 +\\
 +Why do learners sometimes stick to the narrow generalizations [2,3] and sometimes make a wider generalization (category-based) [1]? 
 +\\
 +\\
 In recent work, we (Reeder, Newport, & Aslin, 2009, 2010) In recent work, we (Reeder, Newport, & Aslin, 2009, 2010)
 demonstrated a similar phenomenon—and described some of demonstrated a similar phenomenon—and described some of
Line 122: Line 152:
 B), much like subjects, verbs, and direct objects in sentences B), much like subjects, verbs, and direct objects in sentences
 such as “Bill ate lunch.” Depending on the experiment, the such as “Bill ate lunch.” Depending on the experiment, the
-input included sentences in which all of the words within a +input included sentences in which **all of the words within a 
-particular category occurred in the same contexts (e.g., words+particular category occurred in the same contexts** (e.g., words
 X1, X2, and X3 all occurred after any of the A words and before X1, X2, and X3 all occurred after any of the A words and before
-any of the B words), or the input included only sentences in+any of the B words), or **the input included only sentences in
 which the X words occurred in a limited number of overlapping which the X words occurred in a limited number of overlapping
-A-word or B-word contexts.+A-word or B-word contexts**.
 Adult learners are surprisingly sensitive to these differences. Adult learners are surprisingly sensitive to these differences.
-Our results showed that participants’ tendency to generalize+Our results showed that **//participants’ tendency to generalize
 depended on the precise degree of overlap among word depended on the precise degree of overlap among word
 contexts that they heard in the input, and also on the consistency contexts that they heard in the input, and also on the consistency
 with which a particular A or B word was missing from with which a particular A or B word was missing from
-possible X-word contexts. Adults generalize rules when the+possible X-word contexts//**.  
 +\\ 
 +\\ 
 +**Adults generalize rules when the
 shared contexts are largely the same, with only an occasional shared contexts are largely the same, with only an occasional
 absence of overlap (i.e., a “gap”). However, when the gaps are absence of overlap (i.e., a “gap”). However, when the gaps are
 persistent, adults judge them to be legitimate exceptions to the persistent, adults judge them to be legitimate exceptions to the
-rule and no longer generalize to these contexts. Thus, similar+rule and no longer generalize to these contexts.** 
 +\\ 
 +\\ 
 +//MyNote//: this is a broad description of the observed results, but no explanation as to why this is the case, and no precision in describing: "largely", "persistent" -> What is large enough? When is persistent enough? Why? 
 +\\ 
 +\\ 
 +Thus, similar
 to the results of Gerken (2006), our findings showed that it to the results of Gerken (2006), our findings showed that it
 was the consistency of context cues that led learners to generalize was the consistency of context cues that led learners to generalize
Line 163: Line 202:
 accomplished without instruction, through mere exposure to accomplished without instruction, through mere exposure to
 structured input. structured input.
 +
 +----
 +**Conclusion:**
 +\\
 +Perceptual salience and the patterning of context cues are not
 +the only factors that can influence what learners acquire via a
 +statistical-learning mechanism. An extensive literature in linguistics
 +has argued that languages of the world display a small
 +number of universal patterns—or a few highly common patterns,
 +out of many that are possible—and has suggested that
 +language learners will fail to acquire languages that do not
 +exhibit these regularities (Chomsky, 1965, 1995).
 +Recently, a number of studies using
 +artificial grammars have indeed shown that both children and
 +adults will more readily acquire languages that observe the
 +universal or more typologically common patterns found in
 +natural languages.
 +For example, Hudson Kam and Newport (2005, 2009) and
 +Austin and Newport (2011) presented adults and children with
 +miniature languages containing inconsistent, probabilistically
 +occurring forms (e.g., nouns were followed by the nonsense
 +word ka 67% of the time and by the nonsense word po the
 +remaining 33% of the time). This type of probabilistic variation
 +is not characteristic of natural languages, but it does occur
 +in the speech of nonnative speakers who make grammatical
 +errors. Adult learners in these experiments matched the probabilistic
 +variation they had heard in their input when they produced
 +sentences using the miniature language, but young
 +children formed a regular rule, producing ka virtually all of the
 +time, thereby restoring to the language the type of regularity
 +that is more characteristic of natural languages.
 +\\
 +\\
 +It is not always clear why learners acquire certain types of
 +patterns more easily than others (and why languages therefore
 +more commonly exhibit these patterns). Some word orders
 +place prominent words in more consistent positions across different
 +types of phrases; other patterns are more internally regular
 +or conform better to the left-to-right biases of auditory
 +processing. A full understanding of the principles underlying
 +these learning outcomes awaits further research. What is clear,
 +however, is that statistical learning is not simply a veridical
 +reproduction of the stimulus input. Learning is shaped by a
 +number of constraints on perception and memory, at least
 +some of which may apply not only to languages but also to
 +nonlinguistic patterns.