The upper panel of Figure 1 shows a matrix of frame-level similarity scores between these two utterances where lighter grey represents higher similarity.

All similarity scores are then normalized to the range of [0, l], which yields similarity matrices exemplified in the upper panel of Figure 1.

Given an M -by-N matrix of frame-level similarity scores , the top-left corner is considered the origin, and the bottom-right comer represents an alignment of the last frames in each sequence.

Park-Glass similarity scores by themselves can attribute a high score to distorted paths that, in our context, ultimately leads to too many false-alarm alignments, even after applying the distortion threshold.

MEAD uses a redundancy removal mechanism similar to MMR, but to decide the salience of a sentence to the whole topic, MEAD uses not only its similarity score but also sentence position, e.g., the first sentence of each new story is considered important.

Table 2: Context-sensitive similarity scores (in bold) for the Y slots of four rule applications.

Rather than computing a single context-insensitive rule score, we compute a distinct word-level similarity score for each topic in an LDA model.

At learning time, we compute for each candidate rule a separate, topic-biased, similarity score per each of the topics in the LDA model.

Then, at rule application time, we compute an overall reliability score for the rule by combining the per-topic similarity scores , while biasing the score combination according to the given context of 212.

sim/3m), we compute a topic-biased similarity score for each LDA topic 75, denoted by simt(v, v’ simt(v, v’) is computed by applying

The sense similarity scores are computed by using the vector space model.

Similarity scores are used as additional features of the translation model to improve translation performance.

In Alg2, the similarity score consists of three parts as in Equation (14): sim(Cf“”,C;""c) , sim(Cf‘”,C§""c) , and sim(C§°oc,C:""C) ; where sim(CJf.0°C,C:0“) could be computed by IBM model 1 probabilities simIBM(C;0“,C:OOC) or cosine distance similarity function simCOS(C;OOC,C:W) .

The monolingual similarity scores give it the ability to avoid “dangerous” words, and choose alternatives (such as larger phrase translations) when available.

We then combine the two similarity scores by using both of them as features to see if we could obtain further improvement.

The sense similarity scores are used as feature functions in the translation model.

To calculate the similarity scores of path alignments, we use the sum of word vectors of the words from each path, and calculate the cosine similarity.

For example, the similarity score of the path alignment “OB J (blame) I OB J -ARG(death) m SUB J (cause)OB J -ARG(loss)MOD-ARG(life)” is calculated as the cosine similarity of vectors blame+death and cause+loss+life.

60%, which is fairly high, given our rough estimation of the similarity score .

Aligned paths are evaluated by a similarity score to estimate their likelihood of being paraphrases.

Aligned paths are evaluated by a similarity score , for which we use distributional similarity of the words that appear in the paths (§4.1).

Only path alignments with high similarity scores can be accepted.

We propose an automatic machine translation (MT) evaluation metric that calculates a similarity score (based on precision and recall) of a pair of sentences.

Unlike most metrics, we compute a similarity score between items across the two sentences.

The weights (from the edges) of the resulting graph will then be added to determine the final similarity score between the pair of sentences.

To obtain a single similarity score scores for this sentence pair 3, we simply average the three Fmean scores.

Then, to obtain a single similarity score Sim-score for the entire system corpus, we repeat this process of calculating a scores for each system-reference sentence pair 3, and compute the average over all |S | sentence pairs:

In an n-gram bipartite graph, the similarity score , or the weight 212(6) of the edge 6 connecting a system

Although soft-matching approaches can improve the similarity scores , they are corpus-dependent and match relaxations may be task-specific.

We propose an alternative approach called descending path kernel which gives intuitive similarity scores on comparable structures.

For example, the similarity score between the NPs in Figure l(b) would be zero since the production rule is different (the overall similarity score is above-zero because of matching pre-terminals).

This kernel uses a descending path representation in trees to allow higher similarity scores on partially matching structures, while being simpler and faster than other methods for doing the same.

For the tree kernel KT, subset tree (SST) kernel was applied on each tree representation p. The final similarity score between two instances is the T-weighted sum of the similarities of all representations, combined with the flat feature (FF) similarity as measured by a feature kernel K F (linear or polynomial).

This approach assigns more robust similarity scores (e. g., 78% similarity in the above example) than other soft matching tree kernels, is faster than the partial tree kernel (Moschitti, 2006), and is less ad hoc than the grammar-based convolution kernel (Zhang et al., 2007).

Unlike SST and PTK, once the root category comparison is successfully completed, DPK looks at all paths that go through it and accumulates their similarity scores independent of ordering — in other words, it will ignore the ordering of the children in its pro-

This means, for example, that if the rule production NP —> NN J J DT were ever found in a tree, to DPK it would be indistinguishable from the common production NP —> DT JJ NN, despite having inverted word order, and thus would have a maximal similarity score .

Because topic-specific rules usually have a larger sensitivity score, they can beat general rules when they obtain the same similarity score against the input sentence.

The similarity scores indicate that “deliver X” and “distribute X” are more appropriate to translate the sentence.

The similarity scores are integrated into the standard log-linear model for making translation decisions.

The similarity score of the representation pair (zf, 26) is defined as the cosine similarity of the two vectors:

Since a parallel sentence pair should have the same topic, our goal is to maximize the similarity score between the source sentence and target sentence.

Our algorithm introduces nonaligned signatures (NAS), a cross-lingual word context similarity score that avoids the over-constrained and inefficient nature of alignment-based methods.

We now rank the candidates according to the nonaligned signatures (NAS) similarity score , which assesses the similarity between each candidate’s signature and that of the headword.

3.4 Nonaligned Signatures (NAS) Similarity Scoring

At the heart of our method is the nonaligned signatures (NAS) context similarity score , used for removing incorrect translations using cross-lingual co-occurrences.

The common method for context similarity scoring utilizes some algebraic distance between context vectors, and requires a single alignment of context vectors in one language into the other.

We present the nonaligned signatures (NAS) similarity score for signature and use it to rank these translations.

In this way, the two scores are ‘plugged’ into our method and serve as baselines for our NAS similarity score .

Given a similarity score for all pairs of sentences in the dataset, a regressor is trained on the training set to map the system’s output to the gold standard scores.

Then, for STS 2012, 1,500 pairs were selected and annotated with similarity scores .

0 Pearson correlation: The Pearson correlation between the system’s similarity scores and the human gold-standards.

KB: The knowledge base is a set of lexical and phrasal rules generated from distributional semantics, along with a similarity score for each rule (section 2.6).

where vs_sim is a similarity function that calculates the distributional similarity score between the two lexical predicates.

To produce a final similarity score, we train a regressor to learn the mapping between the two PSL scores and the overall similarity score .

Meyer and Gurevych (2011) showed that automatic alignments between Wiktionary senses and PWN can be established with reasonable accuracy and recall by combining multiple text similarity scores to compare a bag of words based on several pieces of information linked to a WordNet sense with another bag of words obtained from a Wiktionary entry.

We calculated a number of similarity scores , the first two based on similarity in the number of lemmas, calculated using the J accard index:

This development dataset was used to tune refined similarity scores .

Sentence Similarity is the process of computing a similarity score between two sentences.

A subset of 30 pairs is further selected by L106 to render the similarity scores evenly distributed.

The performance of WTMF on CDR is compared with (a) an Information Retrieval model (IR) that is based on surface word matching, (b) an n-gram model (N-gram) that captures phrase overlaps by returning the number of overlapping ngrams as the similarity score of two sentences, (c) LSA that uses svds() function in Matlab, and (d) LDA that uses Gibbs Sampling for inference (Griffiths and Steyvers, 2004).

Using a smaller wm means the similarity score is computed mainly from semantics of the observed words.

This benefits CDR, since it gives more accurate similarity scores for those similar pairs, but not so accurate for dissimilar pairs.

Rex estimates a similarity score for each of the 1,264,827 pairings of comparable terms it finds in the Google 3-grams.

Negating the log of this normalized length yields a corresponding similarity score .

Using the Google n-grams as a source of tacit grouping constructions, we have created a comprehensive lookup table that provides Rex similarity scores for the most common (if often implicit) comparisons.

Comparability is not the same as similarity, and a nonzero similarity score does not mean that two concepts would ever be considered comparable by a human.

We define the similarity score as sim(6di, 661*) = 1 — JSD(6di||6d*).1 Thus, we obtain a vector of similarity scores indexed by the training conversations.

X —> Y added to the search graph, its topic similarity score as follows:

Phrase pairs from the “background conversation” only are assigned a similarity score FX_>y = 0.00.

The candidate answers are scored using a linear interpolation of two cosine similarity scores : one between the entire parent document and question (to model global context), and a second between the answer candidate and question (for local context).6 Because the number of answer candidates is typically large (e.g., equal to the number of paragraphs in the textbook), we return the N top candidates with the highest scores.

If text before or after a marker out to a given sentence range matches the entire text of the question (with a cosine similarity score larger than a threshold), that argument takes on the label QSEG, or OTHER otherwise.

The values of the discourse features are the mean of the similarity scores (e. g., cosine similarity using tfidf weighting) of the two marker arguments and the corresponding question.

Both this overall similarity score , as well as the average pairwise cosine similarity between each word in the question and answer candidate, serve as features.

The distributional similarity scores of the nearest neighbours are associated with the respective target word senses using a WordNet similarity measure, such as those proposed by J iang and Conrath (1997) and Banerjee and Pedersen (2002).

The word senses are ranked based on these similarity scores , and the most frequent sense is selected for the corpus that the distributional similarity thesaurus was trained over.

To compute the similarity between a sense and a topic, we first convert the words in the gloss/definition into a multinomial distribution over words, based on simple maximum likelihood estimation.6 We then calculate the Jensen—Shannon divergence between the multinomial distribution (over words) of the gloss and that of the topic, and convert the divergence value into a similarity score by subtracting it from 1.

The prevalence score for a sense is computed by summing the product of its similarity scores with each topic (i.e.

The first three algorithms produce similarity scores by matching words in the two phrases and the fourth is an edit distance.

The algorithm is essentially bottom-up agglomerative clustering of word senses using a similarity score derived from P(Y|X1, X2).

The similarity score for two senses is defined as:

When computing distributional similarity scores , a template is represented as a feature vector of the CUIs that instantiate its arguments.

Next, we represent each pair of propositional templates with a feature vector of various distributional similarity scores .

A template pair is represented by a feature vector where each coordinate is a different distributional similarity score .

We then generate for any (t1, t2) features that are the 12 distributional similarity scores using all combinations of the dimensions.

In this case, both the definitional and structural similarity scores are treated as equally important and two concepts are aligned if their overall similarity exceeds the middle point of the similarity scale.

If their similarity score exceeds a certain value denoted by 6

Each of these components gets, as its input, a pair of concepts belonging to two different semantic networks and produces a similarity score .

Our present model makes strong assumptions about the independence of similarity scores .

We represent each distinct keyphrase as a vector of similarity scores computed over the set of observed keyphrases; these scores are represented by s in Figure 2, the plate diagram of our model.1 Modeling the similarity matrix rather than the sur-

1We assume that similarity scores are conditionally independent given the keyphrase clustering, though the scores are in fact related.

To address this issue, we develop a manifold model (Belkin et al., 2006) that encourages examples (including both labeled and unlabeled examples) with similar contents to be assigned with similar scores .

Scores are fit so that examples (both labeled and unlabeled) with similar content get similar scores , and scores of labeled examples are close to their labels.

In addition, we also want f to preserve the manifold topology of the dataset, such that similar examples (both labeled and unlabeled) get similar scores .

To integrate two similarity scores , we adopt an average as a composite function.

We finally compute initial similarity scores for all pairs (6, c) where e 6 V6 and c 6 VC, and build the initial similarity matrix R0.

From R”, we finally extract one-to-one matches by using simple greedy approach of three steps: (1) choosing the pair with the highest similarity score ; (2) removing the corresponding row and column from R”; (3) repeating (l) and (2) until the matching score is not less than a threshold 6.

One possible strategy is to compute similarity scores 8(Wfi, we/) between the current source word feature vector Wfi and feature vectors we/Eve for all possible candidates in the target vocabulary.

Following this, we can prune the translation candidate set by keeping only the top candidates 6* according to the similarity scores .

This makes the complexity far worse (in practice) since the dimensionality of the feature vectors d is a much higher value than Computing similarity scores alone (nai'vely) would incur O(|Ve| - d) time which is prohibitively huge since we have to do this for every token in the source language corpus.

All of these high-affinity pairs have a similarity score higher than 0.72.

These two sets of similarity scores are then plotted in a scatter plot, as in Figure 4.

Then, the relation matrix of a bitext is built of similarity scores for the rough translation and the actual translation at sentence level.

In the pre-training stage, Stacked Denoising Auto-encoders are built in an unsupervised layer-wise fashion to discover general concepts encoding d and e. In the supervised fine-tuning stage, the entire network weights are fine-tuned to optimize the similarity score sim(d, e).

The similarity score of (d, 6) pair is defined as the dot product of f (d) and f (6) (Fig.

That is, we raise the similarity score of true pair sim(d, e) and penalize all the rest sim(d, 6,).

For each pair of nodes (u,v) in the graph, we compute the semantic similarity score (using WordNet) between every pair of dependency relation (rel: a, b) in u and v as: s(u,v) = Z WN(a,-,aj) >< WN(b,-,bj),

WN(w,—, wj) is defined as the WordNet similarity score between words 212,- and to]?

For example, the sentences “I adore tennis” and “Everyone likes tennis” convey the same view and should be assigned a higher similarity score as opposed to “I hate tennis”.

Then, for each phrase pair extracted from the training data, we create a vector with features defined in the same way, and calculate its similarity score with the vector representing the dev set.

VSM uses the similarity score between the vec-

To further improve the similarity score , we apply absolute discounting smoothing when calculating the probability distributions p,( f, e).

However, common to all datasets is that similarity scores are given to pairs of words in isolation.

Single-prototype models would give the max similarity score for those pairs, which can be problematic depending on the words’ contexts.

For evaluation, we also compute Spearman correlation between a model’s computed similarity scores and human judgments.

Third, we compared to the entailment classifier with no transitivity constraints (clsf) to see if combining distributional similarity scores improves performance over single measures.

similarity score estimating whether p1 entails p2.

We compute 11 distributional similarity scores for each pair of predicates based on the arguments appearing in the extracted arguments.

Internally, the ranking uses Jensen-Shannon (Lee, 1999) to compute similarity scores between internal representations of seed attributes, on one hand, and each of the candidate attributes, on the other hand.

4) ranking of candidate attributes with respect to each class (e.g., movies), by computing similarity scores between their individual vector representations and the reference vector of the seed attributes.

To this effect, the extraction includes modifications such that only one reference vector is constructed internally from the seed attributes during the third stage, rather one such vector for each class in (Pasca, 2007); and similarity scores are computed cross-class by comparing vector representations of individual candidate attributes against the only reference vector available during the fourth stage, rather than with respect to the reference vector of each class in (Pasca, 2007).