Techniques that compare short text segments using dependency paths (or simply, paths) appear in a wide range of automated language processing applications including question answering (QA).

In this paper, we introduce a flexible notion of paths that describe chains of words on a dependency path .

Dependency paths (or simply, paths) are compared using techniques such as tree edit distance (Punyakanok et al., 2004; Heilman and Smith, 2010), relation probability (Gao et al., 2004) and parse tree alignment (Wang et al., 2007; Park et al., 2011).

Much work on sentence similarity using dependency paths focuses on question answering (QA) where textual inference requires attention to linguistic detail.

In this paper, we explore a flexible application of dependency paths that overcomes this difficulty.

Techniques that compare short text segments using dependency paths are applied to a wide range of automated language processing tasks, including paraphrasing, summarization, entailment detection, QA, machine translation and the evaluation of word, phrase and sentence similarity.

A generic approach uses a matching function to compare a dependency path between any two stemmed terms cc and y in a sentence A with any dependency path between cc and y in sentence B.

The match score for A and B is computed over all dependency paths in A.

o dependency path between a’s head and the predicate

o the set of dependency labels of the predicate’s children 0 dependency path conjoined with the POS tag of a’s head

0 dependency path conjoined with the word cluster of a’s head

second example, for the predicate run, the agent The athlete is not a direct dependent, but is connected via a longer dependency path .

Dependency Paths To capture more relevant context, we developed a second context function as follows.

We scanned the training data for a given task (either the PropBank or the FrameNet domains) for the dependency paths that connected the gold predicates to the gold semantic arguments.

Sentences like this have very long (and thus rare) lexical features, but relatively short dependency paths .

For each sentence we extract a dependency path between each pair of entities.

A dependency path consists of a series of dependencies, directions and words/chunks representing a traversal of the parse.

Part-of-speech tags are not included in the dependency path .

WOE can run in two modes: a CRF extractor (WOEPOS) trained with shallow features like POS tags; a pattern classfier (WOEparse) learned from dependency path patterns.

We show that abstract dependency paths are a highly informative feature when performing unlexicalized extraction.

(Snow et al., 2005) utilize WordNet to learn dependency path patterns for extracting the hypernym relation from text.

However, our results imply that abstracted dependency path features are highly informative for open IE.

WOEparse uses a pattern learner to classify whether the shortest dependency path between two noun phrases indicates a semantic relation.

Despite some evidence that parser-based features have limited utility in IE (Jiang and Zhai, 2007), we hoped dependency paths would improve precision on long sentences.

Shortest Dependency Path as Relation: Unless otherwise noted, WOE uses the Stanford Parser to create dependencies in the “collapsedDepen-dency” format.

We extract dependency paths for each pair of named entities in one sentence.

for words on the dependency paths .

Each entity pair tun and the dependency path which connects them form wit a tuple.

Such patterns could be sequences of lemmas and Part-of-Speech tags, or lexicalized dependency paths .

Whether we use sequences or dependency paths , we will encounter the problem of polysemy.

We perform experiments on New York Times articles and consider lexicalized dependency paths as patterns in our data.

Both DIRT and our approach represent dependency paths using their arguments.

where s and 7“ denote “source” and “receiver” arguments, which are political actor entities in a predefined set 5, t is a timestep (i.e., a 7-day period) derived from the article’s published date, and wpredpath is a textual predicate expressed as a dependency path that typically includes a verb (we use the terms “predicate-pat ” and “verb-pat ” interchangeably).

Verb paths are identified by looking at the shortest dependency path between two mentions in a sentence.

Many of our dependency paths, when traversed from the source to receiver direction, also follow surface order, due to English’s SVO word order.6 Therefore we convert each path to a word sequence and match against the TABARI lexicon—plus a few modifications for differences in infinitives and stemming—and find 528 dependency path matches.

We also create a baseline El-regularized logistic regression that uses normalized dependency path counts as the features (10,457 features).

o For each frame k, draw a multinomial distribution of dependency paths, gbk; N Dir(b / V) (where V is the number of dependency path types).

The vanilla model is capable of inducing frames through dependency path co-occurences, when multiple events occur in a given context.

Figure 5: Dependency path

From this dependency tree, we can extract a dependency path shown in Figure 5, which appears to be a strong clue in knowing that these proteins are mentioned as interacting.

Figure 6: Tree representation of a dependency path

The similarity of two sentences is defined as the bag-of-words similarity of the dependency paths connecting arguments.

o (3) Syntactic features representing the dependency path between two arguments, such as “subj”, “pred”, “modJiprep” and “objprep” (between arguments “antibiotic” and “lyme disease”) in Figure 2.

0 (5) Topic features modeling the words in the dependency path .

In the example given in Figure 2, the dependency path contains the following words: “be”, “standard therapy” and “for”.

We begin by locating terminal nodes, i.e., those which have no incoming edges, depicted as filled circles in Figure 3, and find a dependency (singly linked) path from each terminal node to the root, or a node labeled ‘E’ here, which would give us two paths p1 = ACDE and p2 = BCDE (call them terminating dependency paths , or TDPs).

Later in the paper, we will introduce an approach called the ‘Dependency Path Model’ (DPM) from the previous literature (Section 4), which purports to provide a robust framework for sentence compres-

In what follows, we will describe somewhat in detail a prior approach to sentence compression in Japanese which we call the ”dependency path model,” or DPM.

Dependency path length (DL) refers to the number of (singly linked) dependency relations (or edges) that span two bunsetsa’s.

We define dependency patterns as sub-paths of the shortest dependency path between a concept pair for two reasons.

Shortest dependency path inducement.

From the original dependency tree structure by parsing the selected sentence for each concept pair, we first induce the shortest dependency path with the entitled concept and related concept.

Two ways of extracting patterns have been used: (a) Syntactic, taking the dependency path between the two entities, and (b) Intertext, taking the text between the two.

This context may be a complex structure, such as the dependency path joining the two entities, but it is considered for our purposes as a single term; (e) for each relation r relating 67; with 63-, document Dij is added to collection CT.

The words in each document can be, for example, all the dependency paths that have been observed in the input textual corpus between the two related entities.

Generative model Once these collections are built, we use the generative model from Figure 2 to learn the probability that a dependency path is conveying some relation between the entities it connects.

(2009), we include the notion of verb and noun supports and sections of the dependency path .

This highlights an important advantage of the pipeline trained model: the features can consider any part of the syntax (e. g., arbitrary sub-trees), whereas the joint model is limited to those features over which it can efficiently marginalize (e.g., short dependency paths ).

Even at the expense of no dependency path features, the joint models best pipeline-trained models for state-of-the-art performance in the low-resource setting (§ 4.4).

(2009), who utilize features on syntactic siblings and the dependency path concatenated with the direction of each edge.

We therefore mapped the DIRT Inference rules (Lin and Pantel, 2001), (which consist of pairs of dependency paths ) to TEXTRUNNER relations as follows.

From the parses we extracted all dependency paths between nouns that contain only words present in the TEXTRUNNER relation string.

These dependency paths were then matched against each pair in the DIRT database, and all pairs of associated relations were collected producing about 26,000 inference rules.

entity types, a dependency path and maybe a span word, if g has one.

Each guideline g={gi|i=1,2,3} consists of a pair of semantic types, a dependency path , and optionally a span word and is associated with a particular relation r(g).

Table 3: Performance of a MaxEnt, trained on hand-labeled data using all features (Surdeanu et al., 2011) vs using a subset of two (types of entities, dependency path ), or three (adding a span word) features, and evaluated on the test set.

The syntactic rules correspond to the shortest dependency paths between an opinion word and an extracted mention.

We consider the 10 most frequent dependency paths in the training data.

Example dependency paths include nsubj(opinion, mention), nobj(opinion, mention), and am0d(mention, opinion).