nips nips2007 nips2007-1 knowledge-graph by maker-knowledge-mining

1 nips-2007-A Bayesian Framework for Cross-Situational Word-Learning

Source: pdf

Author: Noah Goodman, Joshua B. Tenenbaum, Michael J. Black

Abstract: For infants, early word learning is a chicken-and-egg problem. One way to learn a word is to observe that it co-occurs with a particular referent across different situations. Another way is to use the social context of an utterance to infer the intended referent of a word. Here we present a Bayesian model of cross-situational word learning, and an extension of this model that also learns which social cues are relevant to determining reference. We test our model on a small corpus of mother-infant interaction and ﬁnd it performs better than competing models. Finally, we show that our model accounts for experimental phenomena including mutual exclusivity, fast-mapping, and generalization from social cues. To understand the difﬁculty of an infant word-learner, imagine walking down the street with a friend who suddenly says “dax blicket philbin na ﬁvy!” while at the same time wagging her elbow. If you knew any of these words you might infer from the syntax of her sentence that blicket is a novel noun, and hence the name of a novel object. At the same time, if you knew that this friend indicated her attention by wagging her elbow at objects, you might infer that she intends to refer to an object in a nearby show window. On the other hand if you already knew that “blicket” meant the object in the window, you might be able to infer these elements of syntax and social cues. Thus, the problem of early word-learning is a classic chicken-and-egg puzzle: in order to learn word meanings, learners must use their knowledge of the rest of language (including rules of syntax, parts of speech, and other word meanings) as well as their knowledge of social situations. But in order to learn about the facts of their language they must ﬁrst learn some words, and in order to determine which cues matter for establishing reference (for instance, pointing and looking at an object but normally not waggling your elbow) they must ﬁrst have a way to know the intended referent in some situations. For theories of language acquisition, there are two common ways out of this dilemma. The ﬁrst involves positing a wide range of innate structures which determine the syntax and categories of a language and which social cues are informative. (Though even when all of these elements are innately determined using them to learn a language from evidence may not be trivial [1].) The other alternative involves bootstrapping: learning some words, then using those words to learn how to learn more. This paper gives a proposal for the second alternative. We ﬁrst present a Bayesian model of how learners could use a statistical strategy—cross-situational word-learning—to learn how words map to objects, independent of syntactic and social cues. We then extend this model to a true bootstrapping situation: using social cues to learn words while using words to learn social cues. Finally, we examine several important phenomena in word learning: mutual exclusivity (the tendency to assign novel words to novel referents), fast-mapping (the ability to assign a novel word in a linguistic context to a novel referent after only a single use), and social generalization (the ability to use social context to learn the referent of a novel word). Without adding additional specialized machinery, we show how these can be explained within our model as the result of domain-general probabilistic inference mechanisms operating over the linguistic domain. 1 Os r, b Is Ws Figure 1: Graphical model describing the generation of words (Ws ) from an intention (Is ) and lexicon ( ), and intention from the objects present in a situation (Os ). The plate indicates multiple copies of the model for different situation/utterance pairs (s). Dotted portions indicate additions to include the generation of social cues Ss from intentions. Ss ∀s 1 The Model Behind each linguistic utterance is a meaning that the speaker intends to communicate. Our model operates by attempting to infer this intended meaning (which we call the intent) on the basis of the utterance itself and observations of the physical and social context. For the purpose of modeling early word learning—which consists primarily of learning words for simple object categories—in our model, we assume that intents are simply groups of objects. To state the model formally, we assume the non-linguistic situation consists of a set Os of objects and that utterances are unordered sets of words Ws 1 . The lexicon is a (many-to-many) map from words to objects, which captures the meaning of those words. (Syntax enters our model only obliquely by different treatment of words depending on whether they are in the lexicon or not—that is, whether they are common nouns or other types of words.) In this setting the speaker’s intention will be captured by a set of objects in the situation to which she intends to refer: Is ⊆ Os . This setup is indicated in the graphical model of Fig. 1. Different situation-utterance pairs Ws , Os are independent given the lexicon , giving: P (Ws |Is , ) · P (Is |Os ). P (W| , O) = s (1) Is We further simplify by assuming that P (Is |Os ) ∝ 1 (which could be reﬁned by adding a more detailed model of the communicative intentions a person is likely to form in different situations). We will assume that words in the utterance are generated independently given the intention and the lexicon and that the length of the utterance is observed. Each word is then generated from the intention set and lexicon by ﬁrst choosing whether the word is a referential word or a non-referential word (from a binomial distribution of weight γ), then, for referential words, choosing which object in the intent it refers to (uniformly). This process gives: P (Ws |Is , ) = (1 − γ)PNR (w| ) + γ w∈Ws x∈Is 1 PR (w|x, ) . |Is | The probability of word w referring to object x is PR (w|x, ) ∝ δx∈ w occurring as a non-referring word is PNR (w| ) ∝ 1 if (w) = ∅, κ otherwise. (w) , (2) and the probability of word (3) (this probability is a distribution over all words in the vocabulary, not just those in lexicon ). The constant κ is a penalty for using a word in the lexicon as a non-referring word—this penalty indirectly enforces a light-weight difference between two different groups of words (parts-of-speech): words that refer and words that do not refer. Because the generative structure of this model exposes the role of speaker’s intentions, it is straightforward to add non-linguistic social cues. We assume that social cues such as pointing are generated 1 Note that, since we ignore word order, the distribution of words in a sentence should be exchangeable given the lexicon and situation. This implies, by de Finetti’s theorem, that they are independent conditioned on a latent state—we assume that the latent state giving rise to words is the intention of the speaker. 2 from the speaker’s intent independently of the linguistic aspects (as shown in the dotted arrows of Fig. 1). With the addition of social cues Ss , Eq. 1 becomes: P (Ws |Is , ) · P (Ss |Is ) · P (Is |Os ). P (W| , O) = s (4) Is We assume that the social cues are a set Si (x) of independent binary (cue present or not) feature values for each object x ∈ Os , which are generated through a noisy-or process: P (Si (x)=1|Is , ri , bi ) = 1 − (1 − bi )(1 − ri )δx∈Is . (5) Here ri is the relevance of cue i, while bi is its base rate. For the model without social cues the posterior probability of a lexicon given a set of situated utterances is: P ( |W, O) ∝ P (W| , O)P ( ). (6) And for the model with social cues the joint posterior over lexicon and cue parameters is: P ( , r, b|W, O) ∝ P (W| , r, b, O)P ( )P (r, b). (7) We take the prior probability of a lexicon to be exponential in its size: P ( ) ∝ e−α| | , and the prior probability of social cue parameters to be uniform. Given the model above and the corpus described below, we found the best lexicon (or lexicon and cue parameters) according to Eq. 6 and 7 by MAP inference using stochastic search2 . 2 Previous work While cross-situational word-learning has been widely discussed in the empirical literature, e.g., [2], there have been relatively few attempts to model this process computationally. Siskind [3] created an ambitious model which used deductive rules to make hypotheses about propositional word meanings their use across situations. This model achieved surprising success in learning word meanings in artiﬁcial corpora, but was extremely complex and relied on the availability of fully coded representations of the meaning of each sentence, making it difﬁcult to extend to empirical corpus data. More recently, Yu and Ballard [4] have used a machine translation model (similar to IBM Translation Model I) to learn word-object association probabilities. In their study, they used a pre-existing corpus of mother-infant interactions and coded the objects present during each utterance (an example from this corpus—illustrated with our own coding scheme—is shown in Fig. 2). They applied their translation model to estimate the probability of an object given a word, creating a table of associations between words and objects. Using this table, they extracted a lexicon (a group of word-object mappings) which was relatively accurate in its guesses about the names of objects that were being talked about. They further extended their model to incorporate prosodic emphasis on words (a useful cue which we will not discuss here) and joint attention on objects. Joint attention was coded by hand, isolating a subset of objects which were attended to by both mother and infant. Their results reﬂected a sizable increase in recall with the use of social cues. 3 Materials and Assessment Methods To test the performance of our model on natural data, we used the Rollins section of the CHILDES corpus[5]. For comparison with the model by Yu and Ballard [4], we chose the ﬁles me03 and di06, each of which consisted of approximately ten minutes of interaction between a mother and a preverbal infant playing with objects found in a box of toys. Because we were not able to obtain the exact corpus Yu and Ballard used, we recoded the objects in the videos and added a coding of social cues co-occurring with each utterance. We annotated each utterance with the set of objects visible to the infant and with a social coding scheme (for an illustrated example, see Figure 2). Our social code included seven features: infants eyes, infants hands, infants mouth, infant touching, mothers hands, mothers eyes, mother touching. For each utterance, this coding created an object by social feature matrix. 2 In order to speed convergence we used a simulated tempering scheme with three temperature chains and a range of data-driven proposals. 3 Figure 2: A still frame from our corpus showing the coding of objects and social cues. We coded all mid-sized objects visible to the infant as well as social information including what both mother and infant were touching and looking at. We evaluated all models based on their coverage of a gold-standard lexicon, computing precision (how many of the word-object mappings in a lexicon were correct relative to the gold-standard), recall (how many of the total correct mappings were found), and their geometric mean, F-score. However, the gold-standard lexicon for word-learning is not obvious. For instance, should it include the mapping between the plural “pigs” or the sound “oink” and the object PIG? Should a goldstandard lexicon include word-object pairings that are correct but were not present in the learning situation? In the results we report, we included those pairings which would be useful for a child to learn (e.g., “oink” → PIG) but not including those pairings which were not observed to co-occur in the corpus (however, modifying these decisions did not affect the qualitative pattern of results). 4 Results For the purpose of comparison, we give scores for several other models on the same corpus. We implemented a range of simple associative models based on co-occurrence frequency, conditional probability (both word given object and object given word), and point-wise mutual information. In each of these models, we computed the relevant statistic across the entire corpus and then created a lexicon by including all word-object pairings for which the association statistic met a threshold value. We additionally implemented a translation model (based on Yu and Ballard [4]). Because Yu and Ballard did not include details on how they evaluated their model, we scored it in the same way as the other associative models, by creating an association matrix based on the scores P (O|W ) (as given in equation (3) in their paper) and then creating a lexicon based on a threshold value. In order to simulate this type of threshold value for our model, we searched for the MAP lexicon over a range of parameters α in our prior (the larger the prior value, the less probable a larger lexicon, thus this manipulation served to create more or less selective lexicons) . Base model. In Figure 3, we plot the precision and the recall for lexicons across a range of prior parameter values for our model and the full range of threshold values for the translation model and two of the simple association models (since results for the conditional probability models were very similar but slightly inferior to the performance of mutual information, we did not include them). For our model, we averaged performance at each threshold value across three runs of 5000 search iterations each. Our model performed better than any of the other models on a number of dimensions (best lexicon shown in Table 1), both achieving the highest F-score and showing a better tradeoff between precision and recall at sub-optimal threshold values. The translation model also performed well, increasing precision as the threshold of association was raised. Surprisingly, standard cooccurrence statistics proved to be relatively ineffective at extracting high-scoring lexicons: at any given threshold value, these models included a very large number of incorrect pairs. Table 1: The best lexicon found by the Bayesian model (α=11, γ=0.2, κ=0.01). baby → book hand → hand bigbird → bird hat → hat on → ring bird → rattle meow → kitty ring → ring 4 birdie → duck moocow → cow sheep → sheep book → book oink → pig 1 Co!occurrence frequency Mutual information Translation model Bayesian model 0.9 0.8 0.7 recall 0.6 0.5 0.4 0.3 F=0.54 F=0.44 F=0.21 F=0.12 0.2 0.1 0 0 0.2 0.4 0.6 precision 0.8 1 Figure 3: Comparison of models on corpus data: we plot model precision vs. recall across a range of threshold values for each model (see text). Unlike standard ROC curves for classiﬁcation tasks, the precision and recall of a lexicon depends on the entire lexicon, and irregularities in the curves reﬂect the small size of the lexicons). One additional virtue of our model over other associative models is its ability to determine which objects the speaker intended to refer to. In Table 2, we give some examples of situations in which the model correctly inferred the objects that the speaker was talking about. Social model. While the addition of social cues did not increase corpus performance above that found in the base model, the lexicons which were found by the social model did have several properties that were not present in the base model. First, the model effectively and quickly converged on the social cues that we found subjectively important in viewing the corpus videos. The two cues which were consistently found relevant across the model were (1) the target of the infant’s gaze and (2) the caregiver’s hand. These data are especially interesting in light of the speculation that infants initially believe their own point of gaze is a good cue to reference, and must learn over the second year that the true cue is the caregiver’s point of gaze, not their own [6]. Second, while the social model did not outperform the base model on the full corpus (where many words were paired with their referents several times), on a smaller corpus (taking every other utterance), the social cue model did slightly outperform a model without social cues (max F-score=0.43 vs. 0.37). Third, the addition of social cues allowed the model to infer the intent of a speaker even in the absence of a word being used. In the right-hand column of Table 2, we give an example of a situation in which the caregiver simply says ”see that?” but from the direction of the infant’s eyes and the location of her hand, the model correctly infers that she is talking about the COW, not either of the other possible referents. This kind of inference might lead the way in allowing infants to learn words like pronouns, which serve pick out an unambiguous focus of attention (one that is so obvious based on social and contextual cues that it does not need to be named). Finally, in the next section we show that the addition of social cues to the model allows correct performance in experimental tests of social generalization which only children older than 18 months can pass, suggesting perhaps that the social model is closer to the strategy used by more mature word learners. Table 2: Intentions inferred by the Bayesian model after having learned a lexicon from the corpus. (IE=Infant’s eyes, CH=Caregiver’s hands). Words Objects Social Cues Inferred intention “look at the moocow” COW GIRL BEAR “see the bear by the rattle?” BEAR RATTLE COW COW BEAR RATTLE 5 “see that?” BEAR RATTLE COW IE & CH→COW COW situation: !7.3, corpus: !631.1, total: !638.4

Reference: text

Summary: the most important sentenses genereted by tfidf model

sentIndex sentText sentNum sentScore

1 One way to learn a word is to observe that it co-occurs with a particular referent across different situations. [sent-6, score-0.385]

2 Another way is to use the social context of an utterance to infer the intended referent of a word. [sent-7, score-0.651]

3 Here we present a Bayesian model of cross-situational word learning, and an extension of this model that also learns which social cues are relevant to determining reference. [sent-8, score-0.932]

4 Finally, we show that our model accounts for experimental phenomena including mutual exclusivity, fast-mapping, and generalization from social cues. [sent-10, score-0.529]

5 If you knew any of these words you might infer from the syntax of her sentence that blicket is a novel noun, and hence the name of a novel object. [sent-13, score-0.426]

6 On the other hand if you already knew that “blicket” meant the object in the window, you might be able to infer these elements of syntax and social cues. [sent-15, score-0.581]

7 Thus, the problem of early word-learning is a classic chicken-and-egg puzzle: in order to learn word meanings, learners must use their knowledge of the rest of language (including rules of syntax, parts of speech, and other word meanings) as well as their knowledge of social situations. [sent-16, score-0.947]

8 The ﬁrst involves positing a wide range of innate structures which determine the syntax and categories of a language and which social cues are informative. [sent-19, score-0.683]

9 We ﬁrst present a Bayesian model of how learners could use a statistical strategy—cross-situational word-learning—to learn how words map to objects, independent of syntactic and social cues. [sent-23, score-0.57]

10 We then extend this model to a true bootstrapping situation: using social cues to learn words while using words to learn social cues. [sent-24, score-1.323]

11 1 Os r, b Is Ws Figure 1: Graphical model describing the generation of words (Ws ) from an intention (Is ) and lexicon ( ), and intention from the objects present in a situation (Os ). [sent-27, score-1.065]

12 Dotted portions indicate additions to include the generation of social cues Ss from intentions. [sent-29, score-0.593]

13 Ss ∀s 1 The Model Behind each linguistic utterance is a meaning that the speaker intends to communicate. [sent-30, score-0.298]

14 Our model operates by attempting to infer this intended meaning (which we call the intent) on the basis of the utterance itself and observations of the physical and social context. [sent-31, score-0.604]

15 For the purpose of modeling early word learning—which consists primarily of learning words for simple object categories—in our model, we assume that intents are simply groups of objects. [sent-32, score-0.419]

16 To state the model formally, we assume the non-linguistic situation consists of a set Os of objects and that utterances are unordered sets of words Ws 1 . [sent-33, score-0.41]

17 The lexicon is a (many-to-many) map from words to objects, which captures the meaning of those words. [sent-34, score-0.521]

18 (Syntax enters our model only obliquely by different treatment of words depending on whether they are in the lexicon or not—that is, whether they are common nouns or other types of words. [sent-35, score-0.557]

19 ) In this setting the speaker’s intention will be captured by a set of objects in the situation to which she intends to refer: Is ⊆ Os . [sent-36, score-0.417]

20 Different situation-utterance pairs Ws , Os are independent given the lexicon , giving: P (Ws |Is , ) · P (Is |Os ). [sent-39, score-0.427]

21 We will assume that words in the utterance are generated independently given the intention and the lexicon and that the length of the utterance is observed. [sent-41, score-0.917]

22 Each word is then generated from the intention set and lexicon by ﬁrst choosing whether the word is a referential word or a non-referential word (from a binomial distribution of weight γ), then, for referential words, choosing which object in the intent it refers to (uniformly). [sent-42, score-1.778]

23 |Is | The probability of word w referring to object x is PR (w|x, ) ∝ δx∈ w occurring as a non-referring word is PNR (w| ) ∝ 1 if (w) = ∅, κ otherwise. [sent-44, score-0.561]

24 (w) , (2) and the probability of word (3) (this probability is a distribution over all words in the vocabulary, not just those in lexicon ). [sent-45, score-0.757]

25 The constant κ is a penalty for using a word in the lexicon as a non-referring word—this penalty indirectly enforces a light-weight difference between two different groups of words (parts-of-speech): words that refer and words that do not refer. [sent-46, score-0.945]

26 Because the generative structure of this model exposes the role of speaker’s intentions, it is straightforward to add non-linguistic social cues. [sent-47, score-0.41]

27 We assume that social cues such as pointing are generated 1 Note that, since we ignore word order, the distribution of words in a sentence should be exchangeable given the lexicon and situation. [sent-48, score-1.415]

28 P (W| , O) = s (4) Is We assume that the social cues are a set Si (x) of independent binary (cue present or not) feature values for each object x ∈ Os , which are generated through a noisy-or process: P (Si (x)=1|Is , ri , bi ) = 1 − (1 − bi )(1 − ri )δx∈Is . [sent-54, score-0.682]

29 For the model without social cues the posterior probability of a lexicon given a set of situated utterances is: P ( |W, O) ∝ P (W| , O)P ( ). [sent-56, score-1.092]

30 (6) And for the model with social cues the joint posterior over lexicon and cue parameters is: P ( , r, b|W, O) ∝ P (W| , r, b, O)P ( )P (r, b). [sent-57, score-1.137]

31 (7) We take the prior probability of a lexicon to be exponential in its size: P ( ) ∝ e−α| | , and the prior probability of social cue parameters to be uniform. [sent-58, score-0.882]

32 Given the model above and the corpus described below, we found the best lexicon (or lexicon and cue parameters) according to Eq. [sent-59, score-1.206]

33 Siskind [3] created an ambitious model which used deductive rules to make hypotheses about propositional word meanings their use across situations. [sent-64, score-0.327]

34 This model achieved surprising success in learning word meanings in artiﬁcial corpora, but was extremely complex and relied on the availability of fully coded representations of the meaning of each sentence, making it difﬁcult to extend to empirical corpus data. [sent-65, score-0.62]

35 In their study, they used a pre-existing corpus of mother-infant interactions and coded the objects present during each utterance (an example from this corpus—illustrated with our own coding scheme—is shown in Fig. [sent-67, score-0.522]

36 They applied their translation model to estimate the probability of an object given a word, creating a table of associations between words and objects. [sent-69, score-0.292]

37 Using this table, they extracted a lexicon (a group of word-object mappings) which was relatively accurate in its guesses about the names of objects that were being talked about. [sent-70, score-0.52]

38 Their results reﬂected a sizable increase in recall with the use of social cues. [sent-73, score-0.399]

39 For comparison with the model by Yu and Ballard [4], we chose the ﬁles me03 and di06, each of which consisted of approximately ten minutes of interaction between a mother and a preverbal infant playing with objects found in a box of toys. [sent-75, score-0.285]

40 Because we were not able to obtain the exact corpus Yu and Ballard used, we recoded the objects in the videos and added a coding of social cues co-occurring with each utterance. [sent-76, score-0.948]

41 We annotated each utterance with the set of objects visible to the infant and with a social coding scheme (for an illustrated example, see Figure 2). [sent-77, score-0.734]

42 Our social code included seven features: infants eyes, infants hands, infants mouth, infant touching, mothers hands, mothers eyes, mother touching. [sent-78, score-0.784]

43 For each utterance, this coding created an object by social feature matrix. [sent-79, score-0.49]

44 3 Figure 2: A still frame from our corpus showing the coding of objects and social cues. [sent-81, score-0.729]

45 We coded all mid-sized objects visible to the infant as well as social information including what both mother and infant were touching and looking at. [sent-82, score-0.79]

46 We evaluated all models based on their coverage of a gold-standard lexicon, computing precision (how many of the word-object mappings in a lexicon were correct relative to the gold-standard), recall (how many of the total correct mappings were found), and their geometric mean, F-score. [sent-83, score-0.629]

47 However, the gold-standard lexicon for word-learning is not obvious. [sent-84, score-0.427]

48 Should a goldstandard lexicon include word-object pairings that are correct but were not present in the learning situation? [sent-86, score-0.475]

49 , “oink” → PIG) but not including those pairings which were not observed to co-occur in the corpus (however, modifying these decisions did not affect the qualitative pattern of results). [sent-89, score-0.283]

50 We implemented a range of simple associative models based on co-occurrence frequency, conditional probability (both word given object and object given word), and point-wise mutual information. [sent-91, score-0.524]

51 In each of these models, we computed the relevant statistic across the entire corpus and then created a lexicon by including all word-object pairings for which the association statistic met a threshold value. [sent-92, score-0.777]

52 In order to simulate this type of threshold value for our model, we searched for the MAP lexicon over a range of parameters α in our prior (the larger the prior value, the less probable a larger lexicon, thus this manipulation served to create more or less selective lexicons) . [sent-95, score-0.463]

53 Our model performed better than any of the other models on a number of dimensions (best lexicon shown in Table 1), both achieving the highest F-score and showing a better tradeoff between precision and recall at sub-optimal threshold values. [sent-99, score-0.559]

54 Table 1: The best lexicon found by the Bayesian model (α=11, γ=0. [sent-102, score-0.463]

55 baby → book hand → hand bigbird → bird hat → hat on → ring bird → rattle meow → kitty ring → ring 4 birdie → duck moocow → cow sheep → sheep book → book oink → pig 1 Co! [sent-105, score-0.453]

56 8 1 Figure 3: Comparison of models on corpus data: we plot model precision vs. [sent-123, score-0.306]

57 Unlike standard ROC curves for classiﬁcation tasks, the precision and recall of a lexicon depends on the entire lexicon, and irregularities in the curves reﬂect the small size of the lexicons). [sent-125, score-0.487]

58 While the addition of social cues did not increase corpus performance above that found in the base model, the lexicons which were found by the social model did have several properties that were not present in the base model. [sent-129, score-1.334]

59 First, the model effectively and quickly converged on the social cues that we found subjectively important in viewing the corpus videos. [sent-130, score-0.864]

60 The two cues which were consistently found relevant across the model were (1) the target of the infant’s gaze and (2) the caregiver’s hand. [sent-131, score-0.286]

61 These data are especially interesting in light of the speculation that infants initially believe their own point of gaze is a good cue to reference, and must learn over the second year that the true cue is the caregiver’s point of gaze, not their own [6]. [sent-132, score-0.325]

62 Second, while the social model did not outperform the base model on the full corpus (where many words were paired with their referents several times), on a smaller corpus (taking every other utterance), the social cue model did slightly outperform a model without social cues (max F-score=0. [sent-133, score-2.158]

63 Third, the addition of social cues allowed the model to infer the intent of a speaker even in the absence of a word being used. [sent-137, score-1.024]

64 This kind of inference might lead the way in allowing infants to learn words like pronouns, which serve pick out an unambiguous focus of attention (one that is so obvious based on social and contextual cues that it does not need to be named). [sent-140, score-0.819]

65 Table 2: Intentions inferred by the Bayesian model after having learned a lexicon from the corpus. [sent-142, score-0.463]

66 Situation score is the log probability of the situation (blue dots represent words and objects) under a lexicon (mappings are red lines). [sent-189, score-0.672]

67 When children as young as sixteen months hear a request for a novel word (e. [sent-192, score-0.381]

68 ) they make a surprising inference: they conclude that the novel word applies to a novel object[7, 8]. [sent-195, score-0.401]

69 This inference is surprising because there seems to be no prima facie reason why children should make it—after all, why shouldnt dax simply be another name for a ball? [sent-196, score-0.51]

70 We use the model to infer the best lexicon for a simple artiﬁcial corpus (similar to that used in [10]). [sent-199, score-0.73]

71 This new situation consists of hearing a novel word (“dax”) and seeing both a familiar object and a novel object (BALL and DAX). [sent-201, score-0.73]

72 We then compare four different lexicons on their coverage of both this situation and the previous corpus: (1) one that learns nothing new from the new situation, (2) one that maps dax to BALL, (3) one that maps “dax” to DAX, and (4) one that maps dax to both. [sent-202, score-1.128]

73 While learning both words produces the best score on the new situation, explaining with high probability why the word “dax” was produced, it performs worst on the rest of the corpus. [sent-204, score-0.33]

74 In particular, it gives a low probability to the coincidence that, while “dax” meant BALL the entire time, the model happened never to hear dax when there was a ball around. [sent-205, score-0.538]

75 In contrast, a lexicon learning no new words scores best on the corpus (because of the prior on smaller lexicons) but has no explanation for why it heard the word “dax” in the new situation. [sent-206, score-1.043]

76 The success of this lexicon (robust across a variety of simulations and parameter settings) suggests that explaining the phenomenon of mutual exclusivity may not require appeals to special principles, either pragmatic or languagespeciﬁc. [sent-209, score-0.677]

77 Instead, the mutual exclusivity phenomenon may come from a general goal: to learn the lexicon which best explains the utterances the learner hears, given their context. [sent-210, score-0.751]

78 3 "modi" MODI DAX "modi" DAX MODI DAX Figure 5: Possible outcomes in a social generalization experiment. [sent-223, score-0.374]

79 The eye-gaze of the speaker (pointing to the MODI) is the only cue which determines that the word “modi” should be mapped to the MODI object; despite this, our model ﬁnds the correct mapping. [sent-224, score-0.432]

80 This label refers to the ability of older children to learn a novel label for a novel object in a well-understood sentence frame after only one or a few exposures and retain it over a signiﬁcant delay. [sent-227, score-0.438]

81 There are two surprising components to this task: ﬁrst, the ability of children to learn from a single exposure, and second, the retention of the word for a long period. [sent-228, score-0.427]

82 Although our model cannot speak to the retention interval, our non-social model predicts that a single, ambiguous situation can give enough evidence to learn a new word. [sent-229, score-0.361]

83 We learn a lexicon for a small artiﬁcial corpus that contains some number of function words, which do not co-occur regularly with any object. [sent-231, score-0.728]

84 We then present the model with a new situation in which there is a novel referent, three words that had been “function words” in our corpus, and one new word (analogous to seeing a novel object, a KOBA, and hearing the utterance “this is a Koba! [sent-232, score-0.791]

85 If it learns nothing, it is penalized on its inability to explain the new situation; if it learns a mapping to a function word, it must explain why this function word was not used referentially in the rest of its experience. [sent-235, score-0.32]

86 Thus, when the other words in the utterance are familiar, our model will learn an appropriate lexical mapping from even a single situation. [sent-236, score-0.35]

87 By adding the ability to learn social cues, our model gains the ability to learn words even in fully ambiguous situations. [sent-238, score-0.68]

88 While 12-month-olds were not able to learn that the word “modi” mapped onto the object that the experimenter looked at, both 18- and 24-month-olds correctly made this inference. [sent-242, score-0.391]

89 While the best explanation of this situation was given by assuming that the word “modi” mapped to both novel objects (bottom right), this alternative was not preferred because it added two mappings to the lexicon rather than one. [sent-244, score-1.025]

90 Of the two remaining options, the mapping of “modi” to the correct object was preferred exclusively on account of the distribution of social cues. [sent-246, score-0.485]

91 Much like the older children in Hollich and colleagues’ experiment, our model was ﬁrst able to learn the relevance of particular social cues over the course of its experience (e. [sent-247, score-0.805]

92 7 6 Conclusions We have presented a Bayesian model of cross-situational world-learning which outperforms both baseline associative models and a more sophisticated translation model on learning from noisy corpus data. [sent-250, score-0.391]

93 The ﬁrst is mutual exclusivity, the tendency to avoid mapping a novel word to a familiar object when a novel object is available. [sent-253, score-0.678]

94 Researchers in the psychological literature have attempted to explain this type of phenomenon in terms of both language-speciﬁc constraints and more general social principles. [sent-254, score-0.455]

95 We have illustrated this by giving an extension to our basic model of social intention, in which social cues independently contribute to establishing the focus of referential intention in a particular situation. [sent-259, score-1.204]

96 A strength of this extension is that the model does not need to know before learning which cues are relevant for establishing referential intention (and indeed, these cues may vary across cultures where pointing is accomplished in different ways). [sent-260, score-0.709]

97 While Yu and Ballard [4] modify their model to incorporate the focus of intention, their social model assumes that the socially-salient objects are externally indicated—it cannot learn what cues signal that focus or their relative weights. [sent-261, score-0.824]

98 Using these learned social cues our model is able to succeed in learning words even when there is no consistent pattern of co-occurrece (either because of a lack of data or because of a truly ambiguous situation). [sent-262, score-0.767]

99 Instead, learners can learn a lexicon consisting only of guesses about the meanings of words. [sent-267, score-0.548]

100 A uniﬁed model of word learning: Integrating statistical and social cues. [sent-285, score-0.646]

similar papers computed by tfidf model

tfidf for this paper:

wordName wordTfidf (topN-words)

[('lexicon', 0.427), ('dax', 0.413), ('social', 0.374), ('word', 0.236), ('corpus', 0.235), ('cues', 0.219), ('situation', 0.151), ('intention', 0.132), ('utterance', 0.132), ('koba', 0.124), ('exclusivity', 0.108), ('infant', 0.108), ('os', 0.098), ('lexicons', 0.096), ('words', 0.094), ('objects', 0.093), ('object', 0.089), ('ball', 0.089), ('referent', 0.083), ('cue', 0.081), ('speaker', 0.079), ('mutual', 0.077), ('cow', 0.077), ('children', 0.074), ('ws', 0.073), ('ballard', 0.072), ('novel', 0.071), ('rattle', 0.069), ('referential', 0.069), ('learn', 0.066), ('infants', 0.066), ('caregiver', 0.055), ('meanings', 0.055), ('syntax', 0.055), ('translation', 0.051), ('intent', 0.048), ('mother', 0.048), ('pairings', 0.048), ('yu', 0.048), ('mappings', 0.047), ('linguistic', 0.046), ('psychological', 0.044), ('ambiguous', 0.044), ('modi', 0.044), ('phenomena', 0.042), ('blicket', 0.041), ('hollich', 0.041), ('intends', 0.041), ('oink', 0.041), ('child', 0.039), ('eyes', 0.039), ('phenomenon', 0.037), ('principles', 0.037), ('bear', 0.037), ('ss', 0.037), ('intentions', 0.036), ('older', 0.036), ('pig', 0.036), ('utterances', 0.036), ('model', 0.036), ('threshold', 0.036), ('coded', 0.035), ('language', 0.035), ('precision', 0.035), ('book', 0.034), ('pointing', 0.034), ('associative', 0.033), ('infer', 0.032), ('sentence', 0.031), ('association', 0.031), ('learns', 0.031), ('gaze', 0.031), ('knew', 0.031), ('acquisition', 0.03), ('intended', 0.03), ('childes', 0.028), ('communicative', 0.028), ('elbow', 0.028), ('markson', 0.028), ('moocow', 0.028), ('mothers', 0.028), ('pnr', 0.028), ('pragmatic', 0.028), ('referents', 0.028), ('retention', 0.028), ('sheep', 0.028), ('wagging', 0.028), ('hands', 0.027), ('coding', 0.027), ('scores', 0.027), ('recall', 0.025), ('heard', 0.024), ('touching', 0.024), ('total', 0.024), ('coverage', 0.024), ('surprising', 0.023), ('familiar', 0.023), ('mapping', 0.022), ('creating', 0.022), ('bird', 0.022)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.99999905 1 nips-2007-A Bayesian Framework for Cross-Situational Word-Learning

Author: Noah Goodman, Joshua B. Tenenbaum, Michael J. Black

2 0.14225958 95 nips-2007-HM-BiTAM: Bilingual Topic Exploration, Word Alignment, and Translation

Author: Bing Zhao, Eric P. Xing

Abstract: We present a novel paradigm for statistical machine translation (SMT), based on a joint modeling of word alignment and the topical aspects underlying bilingual document-pairs, via a hidden Markov Bilingual Topic AdMixture (HM-BiTAM). In this paradigm, parallel sentence-pairs from a parallel document-pair are coupled via a certain semantic-ﬂow, to ensure coherence of topical context in the alignment of mapping words between languages, likelihood-based training of topic-dependent translational lexicons, as well as in the inference of topic representations in each language. The learned HM-BiTAM can not only display topic patterns like methods such as LDA [1], but now for bilingual corpora; it also offers a principled way of inferring optimal translation using document context. Our method integrates the conventional model of HMM — a key component for most of the state-of-the-art SMT systems, with the recently proposed BiTAM model [10]; we report an extensive empirical analysis (in many ways complementary to the description-oriented [10]) of our method in three aspects: bilingual topic representation, word alignment, and translation.

3 0.13319805 2 nips-2007-A Bayesian LDA-based model for semi-supervised part-of-speech tagging

Author: Kristina Toutanova, Mark Johnson

Abstract: We present a novel Bayesian model for semi-supervised part-of-speech tagging. Our model extends the Latent Dirichlet Allocation model and incorporates the intuition that words’ distributions over tags, p(t|w), are sparse. In addition we introduce a model for determining the set of possible tags of a word which captures important dependencies in the ambiguity classes of words. Our model outperforms the best previously proposed model for this task on a standard dataset. 1

4 0.11142117 29 nips-2007-Automatic Generation of Social Tags for Music Recommendation

Author: Douglas Eck, Paul Lamere, Thierry Bertin-mahieux, Stephen Green

Abstract: Social tags are user-generated keywords associated with some resource on the Web. In the case of music, social tags have become an important component of “Web2.0” recommender systems, allowing users to generate playlists based on use-dependent terms such as chill or jogging that have been applied to particular songs. In this paper, we propose a method for predicting these social tags directly from MP3 ﬁles. Using a set of boosted classiﬁers, we map audio features onto social tags collected from the Web. The resulting automatic tags (or autotags) furnish information about music that is otherwise untagged or poorly tagged, allowing for insertion of previously unheard music into a social recommender. This avoids the ”cold-start problem” common in such systems. Autotags can also be used to smooth the tag space from which similarities and recommendations are made by providing a set of comparable baseline tags for all tracks in a recommender system. 1

5 0.10766782 84 nips-2007-Expectation Maximization and Posterior Constraints

Author: Kuzman Ganchev, Ben Taskar, João Gama

Abstract: The expectation maximization (EM) algorithm is a widely used maximum likelihood estimation procedure for statistical models when the values of some of the variables in the model are not observed. Very often, however, our aim is primarily to ﬁnd a model that assigns values to the latent variables that have intended meaning for our data and maximizing expected likelihood only sometimes accomplishes this. Unfortunately, it is typically difﬁcult to add even simple a-priori information about latent variables in graphical models without making the models overly complex or intractable. In this paper, we present an efﬁcient, principled way to inject rich constraints on the posteriors of latent variables into the EM algorithm. Our method can be used to learn tractable graphical models that satisfy additional, otherwise intractable constraints. Focusing on clustering and the alignment problem for statistical machine translation, we show that simple, intuitive posterior constraints can greatly improve the performance over standard baselines and be competitive with more complex, intractable models. 1

6 0.088195145 183 nips-2007-Spatial Latent Dirichlet Allocation

7 0.085745245 9 nips-2007-A Probabilistic Approach to Language Change

8 0.082137719 143 nips-2007-Object Recognition by Scene Alignment

9 0.071750276 197 nips-2007-The Infinite Markov Model

10 0.059471771 114 nips-2007-Learning and using relational theories

11 0.058941983 210 nips-2007-Unconstrained On-line Handwriting Recognition with Recurrent Neural Networks

12 0.057723813 169 nips-2007-Retrieved context and the discovery of semantic structure

13 0.055139195 131 nips-2007-Modeling homophily and stochastic equivalence in symmetric relational data

14 0.04782382 71 nips-2007-Discriminative Keyword Selection Using Support Vector Machines

15 0.047470395 22 nips-2007-Agreement-Based Learning

16 0.045227587 150 nips-2007-Optimal models of sound localization by barn owls

17 0.044873271 189 nips-2007-Supervised Topic Models

18 0.04241563 105 nips-2007-Infinite State Bayes-Nets for Structured Domains

19 0.040966474 51 nips-2007-Comparing Bayesian models for multisensory cue combination without mandatory integration

20 0.037950549 68 nips-2007-Discovering Weakly-Interacting Factors in a Complex Stochastic Process

similar papers computed by lsi model

lsi for this paper:

topicId topicWeight

[(0, -0.115), (1, 0.059), (2, -0.036), (3, -0.2), (4, 0.067), (5, -0.01), (6, 0.053), (7, -0.066), (8, -0.037), (9, -0.024), (10, -0.027), (11, -0.038), (12, -0.07), (13, -0.13), (14, -0.056), (15, -0.162), (16, -0.11), (17, 0.068), (18, 0.015), (19, 0.054), (20, 0.041), (21, -0.002), (22, 0.047), (23, -0.07), (24, -0.124), (25, -0.068), (26, -0.006), (27, -0.025), (28, -0.109), (29, 0.083), (30, 0.103), (31, 0.095), (32, -0.137), (33, 0.004), (34, 0.078), (35, -0.12), (36, 0.016), (37, 0.017), (38, 0.045), (39, -0.084), (40, -0.05), (41, -0.033), (42, -0.021), (43, -0.035), (44, 0.049), (45, 0.014), (46, -0.005), (47, 0.06), (48, 0.052), (49, 0.02)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.96478373 1 nips-2007-A Bayesian Framework for Cross-Situational Word-Learning

Author: Noah Goodman, Joshua B. Tenenbaum, Michael J. Black

2 0.78799552 2 nips-2007-A Bayesian LDA-based model for semi-supervised part-of-speech tagging

Author: Kristina Toutanova, Mark Johnson

3 0.7568571 9 nips-2007-A Probabilistic Approach to Language Change

Author: Alexandre Bouchard-côté, Percy Liang, Dan Klein, Thomas L. Griffiths

Abstract: We present a probabilistic approach to language change in which word forms are represented by phoneme sequences that undergo stochastic edits along the branches of a phylogenetic tree. This framework combines the advantages of the classical comparative method with the robustness of corpus-based probabilistic models. We use this framework to explore the consequences of two different schemes for deﬁning probabilistic models of phonological change, evaluating these schemes by reconstructing ancient word forms of Romance languages. The result is an efﬁcient inference procedure for automatically inferring ancient word forms from modern languages, which can be generalized to support inferences about linguistic phylogenies. 1

4 0.67280883 29 nips-2007-Automatic Generation of Social Tags for Music Recommendation

Author: Douglas Eck, Paul Lamere, Thierry Bertin-mahieux, Stephen Green

5 0.55724156 95 nips-2007-HM-BiTAM: Bilingual Topic Exploration, Word Alignment, and Translation

Author: Bing Zhao, Eric P. Xing

6 0.49885923 84 nips-2007-Expectation Maximization and Posterior Constraints

7 0.46685463 210 nips-2007-Unconstrained On-line Handwriting Recognition with Recurrent Neural Networks

8 0.38048711 150 nips-2007-Optimal models of sound localization by barn owls

9 0.38011578 114 nips-2007-Learning and using relational theories

10 0.35333073 131 nips-2007-Modeling homophily and stochastic equivalence in symmetric relational data

11 0.34183854 71 nips-2007-Discriminative Keyword Selection Using Support Vector Machines

12 0.33435565 183 nips-2007-Spatial Latent Dirichlet Allocation

13 0.32009292 22 nips-2007-Agreement-Based Learning

14 0.30825022 169 nips-2007-Retrieved context and the discovery of semantic structure

15 0.30561608 197 nips-2007-The Infinite Markov Model

16 0.28555113 3 nips-2007-A Bayesian Model of Conditioned Perception

17 0.26787564 105 nips-2007-Infinite State Bayes-Nets for Structured Domains

18 0.26628301 198 nips-2007-The Noisy-Logical Distribution and its Application to Causal Inference

19 0.25642088 143 nips-2007-Object Recognition by Scene Alignment

20 0.24228708 113 nips-2007-Learning Visual Attributes

similar papers computed by lda model

lda for this paper:

topicId topicWeight

[(5, 0.048), (13, 0.036), (16, 0.01), (18, 0.027), (21, 0.053), (29, 0.015), (31, 0.013), (34, 0.023), (35, 0.029), (45, 0.388), (47, 0.082), (83, 0.057), (85, 0.024), (87, 0.054), (90, 0.047)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.76959509 1 nips-2007-A Bayesian Framework for Cross-Situational Word-Learning

Author: Noah Goodman, Joshua B. Tenenbaum, Michael J. Black

2 0.60872239 48 nips-2007-Collective Inference on Markov Models for Modeling Bird Migration

Author: M.a. S. Elmohamed, Dexter Kozen, Daniel R. Sheldon

Abstract: We investigate a family of inference problems on Markov models, where many sample paths are drawn from a Markov chain and partial information is revealed to an observer who attempts to reconstruct the sample paths. We present algorithms and hardness results for several variants of this problem which arise by revealing different information to the observer and imposing different requirements for the reconstruction of sample paths. Our algorithms are analogous to the classical Viterbi algorithm for Hidden Markov Models, which ﬁnds the single most probable sample path given a sequence of observations. Our work is motivated by an important application in ecology: inferring bird migration paths from a large database of observations. 1

3 0.49606973 73 nips-2007-Distributed Inference for Latent Dirichlet Allocation

Author: David Newman, Padhraic Smyth, Max Welling, Arthur U. Asuncion

Abstract: We investigate the problem of learning a widely-used latent-variable model – the Latent Dirichlet Allocation (LDA) or “topic” model – using distributed computation, where each of processors only sees of the total data set. We propose two distributed inference schemes that are motivated from different perspectives. The ﬁrst scheme uses local Gibbs sampling on each processor with periodic updates—it is simple to implement and can be viewed as an approximation to a single processor implementation of Gibbs sampling. The second scheme relies on a hierarchical Bayesian extension of the standard LDA model to directly account for the fact that data are distributed across processors—it has a theoretical guarantee of convergence but is more complex to implement than the approximate method. Using ﬁve real-world text corpora we show that distributed learning works very well for LDA models, i.e., perplexity and precision-recall scores for distributed learning are indistinguishable from those obtained with single-processor learning. Our extensive experimental results include large-scale distributed computation on 1000 virtual processors; and speedup experiments of learning topics in a 100-million word corpus using 16 processors. ¢ ¤ ¦¥£ ¢ ¢

4 0.33625412 59 nips-2007-Continuous Time Particle Filtering for fMRI

Author: Lawrence Murray, Amos J. Storkey

Abstract: We construct a biologically motivated stochastic differential model of the neural and hemodynamic activity underlying the observed Blood Oxygen Level Dependent (BOLD) signal in Functional Magnetic Resonance Imaging (fMRI). The model poses a difﬁcult parameter estimation problem, both theoretically due to the nonlinearity and divergence of the differential system, and computationally due to its time and space complexity. We adapt a particle ﬁlter and smoother to the task, and discuss some of the practical approaches used to tackle the difﬁculties, including use of sparse matrices and parallelisation. Results demonstrate the tractability of the approach in its application to an effective connectivity study. 1

5 0.33019063 169 nips-2007-Retrieved context and the discovery of semantic structure

Author: Vinayak Rao, Marc Howard

Abstract: Semantic memory refers to our knowledge of facts and relationships between concepts. A successful semantic memory depends on inferring relationships between items that are not explicitly taught. Recent mathematical modeling of episodic memory argues that episodic recall relies on retrieval of a gradually-changing representation of temporal context. We show that retrieved context enables the development of a global memory space that reﬂects relationships between all items that have been previously learned. When newly-learned information is integrated into this structure, it is placed in some relationship to all other items, even if that relationship has not been explicitly learned. We demonstrate this effect for global semantic structures shaped topologically as a ring, and as a two-dimensional sheet. We also examined the utility of this learning algorithm for learning a more realistic semantic space by training it on a large pool of synonym pairs. Retrieved context enabled the model to “infer” relationships between synonym pairs that had not yet been presented. 1

6 0.32804841 189 nips-2007-Supervised Topic Models

7 0.32803962 93 nips-2007-GRIFT: A graphical model for inferring visual classification features from human data

8 0.32791236 105 nips-2007-Infinite State Bayes-Nets for Structured Domains

9 0.32445607 153 nips-2007-People Tracking with the Laplacian Eigenmaps Latent Variable Model

10 0.32203457 2 nips-2007-A Bayesian LDA-based model for semi-supervised part-of-speech tagging

11 0.32038656 50 nips-2007-Combined discriminative and generative articulated pose and non-rigid shape estimation

12 0.31998065 138 nips-2007-Near-Maximum Entropy Models for Binary Neural Representations of Natural Images

13 0.31937608 18 nips-2007-A probabilistic model for generating realistic lip movements from speech

14 0.31927413 172 nips-2007-Scene Segmentation with CRFs Learned from Partially Labeled Images

15 0.31889099 95 nips-2007-HM-BiTAM: Bilingual Topic Exploration, Word Alignment, and Translation

16 0.3186236 86 nips-2007-Exponential Family Predictive Representations of State

17 0.31555274 63 nips-2007-Convex Relaxations of Latent Variable Training

18 0.31548488 100 nips-2007-Hippocampal Contributions to Control: The Third Way

19 0.31510994 47 nips-2007-Collapsed Variational Inference for HDP

20 0.31484836 34 nips-2007-Bayesian Policy Learning with Trans-Dimensional MCMC